Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
5757 S. University Ave.
Chicago, IL 60637 Main: 773.702.5599
bfi.uchicago.edu
An A�liated Center of
WORKING PAPER · NO. 2021-107
Energy and Environmental Markets, Industrial Organization, and RegulationRyan Kellogg and Mar ReguantSEPTEMBER 2021
ENERGY AND ENVIRONMENTAL MARKETS, INDUSTRIAL ORGANIZATION, AND REGULATION
Ryan KelloggMar Reguant
September 2021
This manuscript is an invited chapter for the forthcoming volume of the 4th Handbook of Industrial Organization by Elsevier. We thank the editors and four anonymous referees for helpful suggestions while preparing this chapter.
© 2021 by Ryan Kellogg and Mar Reguant. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including © notice, is given to the source.
Energy and Environmental Markets, Industrial Organization, and Regulation Ryan Kellogg and Mar ReguantSeptember 2021JEL No. L0,Q2,Q3,Q4,Q5
ABSTRACT
This paper discusses contributions that industrial organization economists have made to our understanding of energy markets and environmental regulation. We emphasize the substantive contributions of recent papers while also highlighting how this literature has adopted and sometimes augmented theoretical and empirical tools from industrial organization. Many of the topics examined by this literature—especially auctions, investment, productivity and innovation, and regulation—also apply to a variety of settings beyond energy and the environment. We also indicate areas where future research is likely to be fruitful, with an emphasis on how industrial organization economists can help inform energy and environmental policies.
Ryan KelloggUniversity of ChicagoHarris School of Public Policy1307 East 60th StreetChicago, IL 60637and [email protected]
Mar ReguantDepartment of EconomicsNorthwestern University302 Donald P. Jacobs2001 Sheridan RoadEvanston, IL 60208and [email protected]
Contents
1 Introduction 5
2 Extraction of energy resources 82.1 Dynamics of oil production and drilling timing . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.1 The “standard” Hotelling model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.2 Augmenting Hotelling by separating drilling and production . . . . . . . . . . . . . 12
2.1.3 Real options: drilling in the presence of stochastic oil prices . . . . . . . . . . . . . 14
2.1.4 OPEC, market power, and drilling sequencing . . . . . . . . . . . . . . . . . . . . . 16
2.2 Oil and gas mineral leasing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.1 Royalties and primary terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.2 Information, common value auctions, and royalties . . . . . . . . . . . . . . . . . . 22
2.2.3 Uncertainty about the number of bidders . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2.4 Sequential lease auctions and firms’ valuations of neighboring tracts . . . . . . . . . 24
2.2.5 Auctions vs unstructured oil and gas leasing . . . . . . . . . . . . . . . . . . . . . . 25
2.2.6 Reassignment of leases after the initial sale . . . . . . . . . . . . . . . . . . . . . . 26
2.2.7 Opportunities for future work on oil and gas leasing . . . . . . . . . . . . . . . . . 27
2.3 Information spillovers and externalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.3.1 A theoretical framework for free-riding and the incentive to delay exploration: the
“war of attrition” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.3.2 Empirical evidence on cross-firm information spillovers and strategic delay . . . . . 31
2.3.3 Exploration spillovers across tracts governed by different extraction policies . . . . . 33
2.4 Productivity, innovation, and learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.4.1 Learning about the production function and optimal input choices . . . . . . . . . . 34
2.4.2 Learning where to drill . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.4.3 Productivity in vertical relationships . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.4.4 Lessons learned and paths for future work on oil and gas productivity . . . . . . . . 39
2.5 Environmental regulation of resource extraction and transportation . . . . . . . . . . . . . . 39
2.5.1 Regulation of environmental damage at production sites and of site decommissioning 39
2.5.2 Regulation of emissions from hydrocarbon transportation . . . . . . . . . . . . . . . 42
2.5.3 Interactions between downstream environmental regulations and market power in
resource transportation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.5.4 Opportunities for future work on impacts of environmental regulation on fossil fuel
extraction and transportation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3 Personal transportation, energy use, and environmental regulation 453.1 Estimating consumers’ demand for fuel economy, and implications for fuel economy policy . 45
3.1.1 Identifying consumers’ valuation of fuel costs from used vehicle prices . . . . . . . 47
3.1.2 Consumer valuation of fuel costs for new vehicles . . . . . . . . . . . . . . . . . . 49
2
3.1.3 Lessons learned and paths forward for research on consumers’ valuation of fuel
economy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.2 Economic impacts of, and firms’ responses to, fuel economy standards . . . . . . . . . . . . 52
3.2.1 Fuel economy standards and automakers’ pricing and fleet mix decisions . . . . . . 52
3.2.2 Fuel economy standards and vehicle attributes . . . . . . . . . . . . . . . . . . . . . 54
3.2.3 Attribute-based fuel economy standards . . . . . . . . . . . . . . . . . . . . . . . . 56
3.2.4 Gaming of fuel economy standards . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.2.5 Lessons learned on fuel economy standards, and paths for future work . . . . . . . . 57
3.3 Industrial organization and vehicles’ emissions of local air pollutants . . . . . . . . . . . . . 58
3.4 Consumers’ fuel search behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.5 Markets for EVs and EV charging stations . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.5.1 Indirect network effects and EV incentive policies . . . . . . . . . . . . . . . . . . 64
3.5.2 Compatibility between charging networks . . . . . . . . . . . . . . . . . . . . . . . 65
3.5.3 Paths for future research on EVs and EV charging . . . . . . . . . . . . . . . . . . 66
4 Electricity markets 674.1 The restructuring of electricity markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.1.1 Aggregate impacts of restructuring . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.1.2 Plant-level evidence of the impacts of restructuring . . . . . . . . . . . . . . . . . . 69
4.1.3 Natural monopoly regulation in distribution . . . . . . . . . . . . . . . . . . . . . . 69
4.2 Market power in wholesale electricity markets . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.2.1 Estimating market power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.2.2 Taking electricity auctions to heart . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.2.3 FOC approach and pass-through analysis . . . . . . . . . . . . . . . . . . . . . . . 76
4.2.4 Market power and dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.2.5 Sequential markets and arbitrage . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.2.6 Modeling transmission and market power . . . . . . . . . . . . . . . . . . . . . . . 79
4.3 Renewable power and the energy transition . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.3.1 Estimating the environmental impact of renewables . . . . . . . . . . . . . . . . . . 81
4.3.2 Estimating the market impacts of renewables . . . . . . . . . . . . . . . . . . . . . 82
4.3.3 Renewables and learning-by-doing . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.3.4 Renewables and demand-side dynamics . . . . . . . . . . . . . . . . . . . . . . . . 85
4.4 Electricity demand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.4.1 Estimating the elasticity of electricity demand . . . . . . . . . . . . . . . . . . . . . 87
4.4.2 Experimental evidence on demand response . . . . . . . . . . . . . . . . . . . . . . 89
4.4.3 Competition in retail electricity markets . . . . . . . . . . . . . . . . . . . . . . . . 90
4.5 Cap-and-trade regulation in electricity markets . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.5.1 Emissions markets in the US electricity sector . . . . . . . . . . . . . . . . . . . . . 92
4.5.2 Interactions between cap-and-trade regulation and regulatory regime . . . . . . . . . 94
3
5 Environmental regulation of energy-intensive industries and natural resources 955.1 Environmental regulation in manufacturing and resource-intensive sectors . . . . . . . . . . 96
5.1.1 Climate regulation and leakage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.1.2 Leakage in dynamic models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.2 Imperfect monitoring and the enforcement of regulations . . . . . . . . . . . . . . . . . . . 98
5.2.1 Evidence of cheating in environmental settings . . . . . . . . . . . . . . . . . . . . 98
5.2.2 Structural models of enforcement . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.3 Regulation of water markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.3.1 Water use and adaptation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
5.3.2 Studying formal water markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6 Concluding remarks 103
4
1 Introduction
Since the 1990s, growing concerns about energy access, the costs of energy supply, and the environmental
damages associated with fossil fuels have led to dramatic changes to energy and environmental markets and
regulation. Examples include the restructuring of the electricity industry away from vertically integrated
utilities and towards independent power producers that compete in wholesale power auctions, introduction
of cap-and-trade markets for greenhouse gases (GHGs) and local air pollutants such as NOx, tightening of
fuel economy standards, and adoption of policies such as renewable portfolio standards to promote zero-
emission energy sources. These policies have been enacted with varying stringency across jurisdictions over
time, and it is clear that policy change will continue into the future, especially in the area of climate policy.
This evolution of regulation and policy has taken place concurrently with remarkable innovation and
productivity improvements in energy production. For fossil fuels, the U.S. shale oil and gas revolution that
started in the mid-2000s was driven by unprecedented productivity growth that led to booms in mineral
leasing, oil and gas production, and crude-by-rail transportation. And for renewables, the costs of wind and
solar electric generation have declined markedly in recent years, such that in some areas they can compete
with fossil fuel based generation on the merits.
Changes in energy and environmental technology, industry structure, and regulation are likely to be even
more dramatic in the years to come with the pressing need to reduce greenhouse gases (GHG) emissions.
The sectors covered in this handbook chapter, either as inputs or as direct outputs, are responsible for over
70% of GHG emissions contributing to global warming. Learning about these sectors is therefore crucial
for any researcher interested in climate change economics.
Understanding the changes in energy supply, energy markets, and regulations—in both a positive and
normative sense—requires addressing questions that are well-suited for the theoretical and empirical tools
from industrial organization. For instance, wholesale electricity markets take the form of high-frequency
multi-unit auctions, vehicle markets can be characterized by differentiated Bertrand competition, extractors
of exhaustible resources face dynamic problems, and energy distribution firms are regulated natural monop-
olies. These market structures and forms of regulation are all areas where IO economists hold expertise.
Our main goal for this chapter is to illustrate how ideas and tools from IO have been marshalled to create
insights into understanding energy and environmental economics and policy.
Our second goal in this chapter is to highlight areas where insights from energy and environmental
markets are likely to be relevant for other settings of interest to IO economists, and cases where methods
developed to address a particular question might have broader applicability. A convenient feature of energy
and natural resource intensive industries for IO researchers is that they tend to be government regulated or
at least government monitored, so that detailed administrative data are available on inputs, outputs, prices,
costs, and investment. These data enable researchers to address broad IO questions that would be challenging
if not impossible to answer credibly using data from other industries.
Given our primary goal of discussing how ideas and tools from IO have helped answer pressing, policy-
relevant questions about energy and environmental markets and regulation, we have organized this chapter
by substantive topic area. We begin in section 2 by discussing IO economists’ contributions to our under-
standing of markets and regulatory policies in the primary energy resource extraction sector. Section 3 then
5
discusses markets and policies for personal transportation, and section 4 discusses the electricity sector.
Given the growing importance of environmental policies—especially policies targeting GHG emissions—
for these markets, each of these three sections emphasizes contributions that speak, directly or indirectly, to
the impacts of such policies. Then in section 5 we discuss environmental regulation and enforcement more
broadly, including regulation of energy-intensive manufacturing industries and markets for water resources.
In each section of this chapter, most of our writing is dedicated to discussing the contributions of IO
economists to the issues at hand, but we also take time as needed to provide industry background that is
often a prerequisite to conducting research in these areas. Our discussions also include subsections that
are forward-looking: what questions remain unanswered, and how might IO economists make contributions
towards answering them? We then conclude the chapter in section 6 by summarizing what we see as the
most promising areas for future contributions from IO to energy and environmental economics and policy,
with an emphasis on research that helps inform policies aimed at accelerating a clean energy transition.
Our organizational approach will be most useful for researchers and students who are interested in
understanding a particular substantive area. For instance, someone who is interested in fuel economy policy
can proceed directly to that section to understand what the compelling research questions have been, how IO
economists have contributed to those questions thus far, and what important unanswered questions remain.
This organization is, though, perhaps less well-suited to those who are interested in applications of specific
models or methods, such as auction models or productivity estimation methods.
Methods cross-walk. To help guide readers interested in particular methods and tools, rather than topical
areas, we conclude this introduction with the following “cross-walk” that indicates where readers can find
applications of—and in many cases, innovations to—models and tools that are broadly used across IO.
1. Auction models have been essential for understanding both the leasing of primary energy resources
and the economics of wholesale electricity markets.
(a) Section 2.2 discusses oil and gas leasing, which has been a long-standing setting for studies
of auction models (Hendricks and Porter, 2007). This section includes discussions of auctions
with contingent payments (oil and gas royalties), implications of bidders’ uncertainty regarding
the number of competitors, sequential versus bundled multi-unit auctions, comparisons between
auctions and unstructured negotiations, and implications of post-auction secondary markets.
(b) Section 4.2 discusses how the study of electricity markets has contributed to the advancing of our
understanding of multi-unit uniform price auctions. Electricity markets provide a unique setting
in which auctions happen very frequently, the players are well-known, and their cost structure
and objective function are well-understood.
2. Productivity estimation has been central to understanding the dramatic recent reductions in the cost
of producing shale oil and gas and of generating electricity from wind and solar resources. It has also
been important for evaluating the consequences of electricity industry restructuring.
6
(a) Section 2.4 discusses how research on the shale oil and gas industry has provided evidence for
forms of productivity improvements that have previously not garnered attention from the litera-
ture on productivity and innovation, including learning about optimal input selection, improving
the selection of which projects to complete, and learning how to work more efficiently with other
firms in the vertical supply chain.
(b) Production function estimation has also been used in the context of electricity markets to assess
the productivity impacts of regulatory reform, discussed in section 4.1, and reductions in the
costs of renewable power generation, discussed in section 4.3.
(c) Section 5.3.2 discusses productivity analysis applied to agricultural production to quantify the
misallocation of water.
3. Differentiated product models have been widely used in work on transportation and the environment
to understand automobile consumers’ demand for fuel economy (section 3.1), automakers’ responses
to fuel economy standards (section 3.2), the economics of regulations on vehicles’ emissions of local
pollutants (section 3.3), and markets for electric vehicles (section 3.5). This work frequently augments
standard Berry et al. (1995) style models of automobile markets to include vehicle characteristics
related to environmental performance and constraints imposed by environmental regulations.
4. Models of consumer search have been used to study how drivers search for retail gasoline and how
households search for retail electricity providers.
(a) Section 3.4 discusses how data from retail gasoline markets have been used to empirically test
models of consumer search. This work has taken advantage of high-frequency spatial data on
retail prices and on search intensity itself.
(b) Section 4.4 discusses how consumer search models have been used to understand competition
in liberalized electricity markets.
5. Single-agent dynamic models are key to modeling the dynamic aspects in energy and environmental
markets.
(a) Dynamic models have long been used, dating back to Hotelling (1931), to understand extraction
of exhaustible resources. Theoretical and empirical applications of these models to the oil and
gas industry are discussed throughout section 2 and especially in section 2.1.
(b) Single-agent models have also been used extensively to model the behavior of electricity market
participants, and discussions of such applications are provided in sections 4.3 and 4.4.
(c) Section 5.2.2 discusses recent developments in structural models of regulation enforcement and
monitoring that also use these dynamic methods.
6. Models of dynamic games are often important because energy industries are often characterized by
long-lived capital investments and oligopolistic competition.
7
(a) Section 2.3 discusses how models of dynamic games have helped build an understanding of
firms’ oil and gas exploration behavior. Even though the large number of private oil and gas
firms typically lends itself to modeling these firms using single-agent dynamic models, in local
exploration contexts there can be important information spillovers across just a small set of firms,
implying that models involving dynamic games are required to understand investment behavior.
(b) Section 5.1.2 shows how current state-of-the-art dynamic games modeling can be applied to
understand environmental leakage applications, and section 4.3.3 discusses the importance of
dynamics and learning-by-doing in the context of renewable technologies.
7. Natural monopoly regulation Many energy sectors have been traditionally regulated (and some still
are) as natural monopolies, making it one of the leading applications of theoretical and empirical
work in this area. In section 4.1, we discuss work examining the regulation of natural monopolies in
the distribution of electricity and natural gas. Section 4.5 discusses how natural monopoly regulation
interacts with environmental regulation.
8. Network goods. Electric vehicles and electric vehicle charging stations are subject to indirect net-
work effects, and we discuss how network goods models have been applied to empirically study this
industry in section 3.5.
9. Vertical relationships appear throughout the energy industry:
(a) Section 2.4 discusses mechanisms by which repeated interactions between well-matched oil
producers and drilling rigs can improve productivity.
(b) Section 2.5 discusses a variety of ways in which vertical relationships between monopoly or
near-monopoly railroads and coal-fired electric generators can create challenges for regulation
of these generators’ pollution emissions.
(c) Section 4.2 discusses the importance of vertical positions in the determination of incentives to
exercise market power in the electricity sector.
2 Extraction of energy resources
This section discusses the industrial organization of primary energy resource extraction. This topic has re-
cently experienced a surge in research, which we see as stemming from two high-level motivations: (1)
topical importance that is driven by both the U.S. shale oil and gas boom and concerns about the environ-
mental consequences of oil and gas production; and (2) unparalleled availability of data on investments in
individual oil wells and on lease contracts between firms and mineral owners. These data can be used to
answer questions about topics such as firms’ investment behavior, oil and gas lease auctions and allocations,
productivity growth, and impacts of environmental regulation.
The oil and gas industry is one of the largest sectors of the economy, has long been thought to be
entwined with macroeconomic outcomes, and can experience dramatic short-run and long-run fluctuations.
In 2019, total world oil production was 95 million barrels per day, worth $6.1 billion per day ($2.2 trillion
8
per year) at 2019 average prices.1 These large magnitudes, combined with oil price volatility arising from
the inelasticity of short-run global demand and supply for oil, has long led economists and policy analysts
to worry about macroeconomic impacts from changes in oil supply (Hamilton, 2013; Kilian and Vigfusson,
2017; Baumeister and Hamilton, 2019). Interest in understanding oil supply has only increased since the
U.S. shale oil and gas boom began in the mid-2000s, more than doubling U.S. oil and gas production and
leading the U.S. to become the world’s largest oil producer in 2019.2
In addition, the negative environmental externalities associated with fossil fuel extraction have increas-
ingly become a target for policy intervention. The GHG emissions associated with the ultimate consumption
of extracted fuels have of course garnered a great deal of attention, but on top of that fossil fuel industries
also generate both local and GHG emissions during the fuels’ extraction and transportation. Industrial or-
ganization has a great deal to contribute to our understanding of the efficacy of regulations that target these
externalities.
Beyond the oil and gas sector’s substantive importance, industrial organization economists have been
drawn to it because the wealth of available data—especially in the U.S.—make it an outstanding setting to
study IO questions about firms’ investment decisions and the economics of contingent contracts between as-
set sellers and well-informed buyers. Every oil and gas well in the U.S., whether drilled on public or private
land, has a publicly-recorded drilling permit and completion report associated with it that describes who
operates it, precisely where it was drilled and to what depth, when drilling began, and when the well was
completed. In many cases, researchers can also observe drilling inputs and the identities of contractors who
worked on the well. Moreover, wells’ monthly production is also observable because it, too, is recorded
by state agencies. The U.S. oil and gas industry therefore offers a unique opportunity to observe a large
number of discrete investments and the return on those investments. For instance, in 2019 approximately
17,000 oil and gas wells were drilled in the U.S. by hundreds if not thousands of firms.3 This granularity of
investment data is difficult to find in almost any other setting. On top of that, a wealth of data are available
regarding leases between mineral owners and extraction firms. For U.S. state and federal minerals, these
leases are typically auctioned, and available data often include information on both winning and losing bids.
Private mineral leases are instead typically negotiated, but because these leases must be publicly recorded,
data are usually available about major lease terms such as the royalty and primary term (though not the
bonus). Together, these data enable researchers to answer questions about how firms compete to acquire de-
velopment contracts, and about how alternative allocation mechanisms affect the owner’s revenues, resource
development, and total surplus.
While this section of the chapter is titled “Extraction of energy resources”, it will follow the recent liter-
ature by focusing the vast majority of its attention on the oil and gas sector. The subsection on environmental
regulation will, however, also discuss coal in the context of downstream emissions regulations and exercise
of market power by railroads. Industrial organization research on coal extraction itself has unfortunately1The 95 million bbl/d figure is from BP (2020), page 16. Valuations are based on the 2019 average Brent crude price of
$64.30/bbl, from the EIA at https://www.eia.gov/dnav/pet/pet_pri_spt_s1_a.htm.2See https://www.eia.gov/dnav/pet/hist/LeafHandler.ashx?n=PET&s=MCRFPUS2&f=M for U.S. oil
production data.3Per Patel and Geary (2020), approximately 1,400 new oil and gas wells were drilled per month in the U.S. in 2019.
9
lagged behind that studying the oil and gas sector, despite coal’s large role in the GHG emissions inventory
and despite a wealth of available data on U.S. coal extraction from agencies such as the U.S. Geological
Survey, Energy Information Administration (EIA), and the Mine Safety and Health Administration.
In terms of methods, tools for simulating and estimating models of dynamic behavior feature promi-
nently in the research we discuss in this section. IO research on resource extraction also draws heavily on
tools for simulation and estimation of auctions, as we discuss in subsection 2.2, and productivity estimation,
as we discuss in subsection 2.4. In general, our discussion of research in this area will emphasize papers’
substantive contributions rather than describe the methods used in detail, in part because the methods them-
selves are discussed extensively in other chapters in this volume.
2.1 Dynamics of oil production and drilling timing
Since Hotelling (1931), economists have recognized that the question of how to optimally produce natural
resources out of an initial stock is inherently a dynamic problem. In most cases, the relevant question is not
whether or not to extract the reserves at all, but instead how much to extract today and how much to leave
for tomorrow. Hotelling’s famous 1931 article presents a prescriptive model of the optimal extraction path
(or, equivalently, the extraction path that would prevail in competitive equilibrium with forward-looking
extractors), culminating in what has become known as Hotelling’s Rule that the difference between the
resource’s price and its marginal extraction cost should rise at the rate of interest. In this sub-section, we
begin by discussing the core intuition and model that underlies this result. We then highlight the empirical
shortcomings of overly-simple Hotelling-style models when applied to oil and gas extraction, and discuss
a series of recent papers in industrial organization that augment the basic version of the Hotelling model
with important real-world features of firms’ extraction problems: capacity constraints on production from
drilled wells, exercise of market power by Organization of Petroleum Exporting Country (OPEC) producers,
and uncertainty about future oil and gas prices. These model augmentations yield vast improvements in the
ability of Hotelling-style models to explain observed market behavior, raising the prospects that such models
can be used to help evaluate the likely consequences of policies affecting the oil and gas industry.
2.1.1 The “standard” Hotelling model
In the standard Hotelling (1931) style model, as commonly taught in graduate resource economics courses,
the central planner (or equivalently, a price-taking, resource-owning firm) is assessing how to optimally
extract an exhaustible resource, given static and deterministic functions for flow utility and extraction costs.
Consider a continuous time version of such a model, letting yt ≥ 0 denote the extraction rate at time t, and
letting xt ≥ 0 denote the remaining stock at t. Utility is given by u(yt), and the extraction cost is c(yt),
such that the difference u(yt) − c(yt) is strictly increasing at yt = 0 and strictly concave. The initial stock
x0 = S > 0.
Given a discount rate r > 0, the planner’s problem is given by:
max{yt}
∫ ∞0
(u(yt)− c(yt)) e−rtdt, s.t. xt = −yt, yt ≥ 0, xt ≥ 0,
10
where xt denotes the rate of change of the stock xt. Letting µt denote the current-time shadow value of the
resource stock, the current-value Hamiltonian and first-order conditions (FOCs) are then:
H = u(yt)− c(yt)− µtyt (1)
FOCyt : u′(yt)− c′(yt)− µt ≤ 0, yt ≥ 0, yt(u′(yt)− c′(yt)− µt
)= 0 (2)
FOCxt : µt = rµt (3)
TVC : limt→∞
xt ≥ 0, limt→∞
µte−rt ≥ 0, lim
t→∞xtµte
−rt = 0. (4)
FOCyt and FOCxt together tell us that on the optimal extraction path, the difference between marginal
utility (i.e. price) and marginal cost must be rising at the interest rate r so long as the extraction rate is strictly
positive. This result is commonly known as the Hotelling Rule. In a competitive equilibrium setting, the
rule relates to an intuitive indifference condition. In order to spread out extraction continuously over time,
extraction firms must be indifferent about extracting their resource across time. This indifference condition
can only hold if the marginal return to extraction is increasing over time at the interest rate, thereby holding
the present value constant.
The transversality condition (TVC) then requires that either the resource be completely exhausted in
the limit as t → ∞, or alternatively that the resource be worthless on the margin in the limit. In most
practical applications with an infinite horizon it is the former condition that will hold, and the TVC then
puts a boundary on the problem and determines the level of the initial extraction rate.
The standard Hotelling Rule prediction is strong and has been shown to be at odds with most natural
resource price data (and, in some studies, implied data on in-situ shadow values). See reviews of these
studies in Krautkraemer (1998) and Slade and Thille (2009). In the context of crude oil prices, figure 1
shows that the time series of U.S. price of crude oil exhibits no strong upward trend over the long run.
One clear gap between the simple Hotelling model presented above and real-world data is that the model
is fully deterministic, whereas real-world demand (u′(yt)) and marginal extraction costs (c′(yt)) of crude
oil are stochastic. Incorporating demand and cost shocks into the simple Hotelling model would naturally
lead the model to predict price volatility (particularly if demand and marginal cost are inelastic in yt), but
Hotelling’s Rule would still hold in expectation. That is, the model would continue to predict that rational
agents at any time t should forecast that the difference between the expected future price and marginal
extraction cost of oil will increase at the interest rate r.
The bottom panel of figure 1 shows that even the expectations version of Hotelling’s Rule fails to hold
in crude oil markets. Using futures prices as a proxy for expected futures prices,4 this panel shows the
evolution over time of the percentage difference between the 12-month oil futures price and the front-month
futures price. A positive value indicates that the market is in contango and that the price is expected to
increase, whereas a negative value indicates that the market is backwardated and that prices are expected to4As explained in Pindyck (2001), futures prices will not be equivalent to expected future prices if the risk-adjusted discount
rate that applies to investments in crude oil stocks differs from the risk-free rate (e.g., if the CAPM beta for crude oil differs fromzero). Even if futures prices are systematically biased away from expected future prices for this reason, the variation in futurescurves between backwardation and contango is not consistent with Hotelling. Also, Anderson et al. (2018) estimates a CAPM betaof nearly zero for West Texas Intermediate crude oil using data from April 1983 through April 2015.
11
Figure 1: Crude oil front-month prices and futures curve growth rates
2040
6080
100
120
140
Fron
t mon
th p
rice
in $
per
bbl
Jan 1990 Jan 1995 Jan 2000 Jan 2005 Jan 2010 Jan 2015 Jan 2020
West TX Intermediate front-month crude oil prices
-100
-50
050
100
Ann
ualiz
ed ra
te o
f pric
e in
crea
se in
%
Jan 1990 Jan 1995 Jan 2000 Jan 2005 Jan 2010 Jan 2015 Jan 2020
Crude futures rate of price increase,from front-month to four-month contract
Note: Futures prices shown are NYMEX West Texas Intermediate crude for delivery to Cushing, Oklahoma.
Data source: https://www.eia.gov/dnav/pet/pet_pri_fut_s1_d.htm. The front-month price
is for delivery in the upcoming month. We compute the annualized rate of price increase as the difference
between the futures price for delivery four months out and the front-month price, divided by the front month
price, times 4 (and converted to percent by multiplying by 100).
fall. The long periods in which oil futures are backwardated, the fluctuations between backwardation and
contango, and the periods of extreme contango when the oil price is expected to rise more quickly than any
plausible discount rate are inconsistent with a Hotelling-style expectation of steadily increasing resource
prices.
2.1.2 Augmenting Hotelling by separating drilling and production
As noted in Slade and Thille (2009), the theoretical and empirical literature on Hotelling-style models had
largely fizzled by the year 2000. Even though there existed some theoretical work such as Pindyck (1978)
that could yield initial periods of decreasing price expectations by requiring initial investments in reserves,
the empirical literature that focused on explaining price time series struggled to fit these models to the data.
Anderson et al. (2018) attempts to address the shortcomings of Hotelling-style models—while retaining
the core dynamic optimization intuition—in the context of oil and gas by focusing attention on micro-
level development and extraction decisions rather than macro-level price time series. The paper begins by
examining lease-level administrative oil production data from Texas. Using data from 16,159 oil leases that
12
did not experience any new well investment after 1990, the paper finds that production from established,
drilled wells is almost completely unresponsive to oil price shocks and instead smoothly falls over time
along a roughly exponential decline curve. This finding prevails despite large oil price swings during the
paper’s 1990–2007 sample and a period in 1998–1999 when 12-month futures prices exceeded the front-
month price by more than 20%. This production behavior violates the predictions of standard Hotelling-style
models.
In contrast, when Anderson et al. (2018) studies data on drilling of new wells, it finds a strong response
of drilling to oil price shocks (the estimated elasticity of drilling to the oil price is 0.73). In addition, the
marginal cost of drilling, measured using data on the rental rate for drilling rigs, positively co-varies with
oil prices. Taken together with the initial finding that oil production from drilled wells is not responsive to
prices, these empirical results signal a need to reformulate Hotelling models in a way that treats production
decisions differently from drilling decisions.
The core innovation that Anderson et al. (2018) uses to explain these results is to follow the petroleum
geology engineering literature by modeling the production rate from drilled wells as subject to a capacity
constraint that is proportional to the remaining reserves available to the well. The idea is that the motive
force that propels oil through the rock, to the wellbore, and up the wellbore is the underground pressure of
the reservoir fluids, and as oil is produced that pressure declines.5 Anderson et al. (2018) then assumes that
the marginal production cost is zero for production rates that are less than this hard constraint. Together, this
additional structure on the extraction problem can deliver the empirical result that firms’ production from
existing wells follows a decline curve and is not responsive to price shocks.
To formally show this result, Anderson et al. (2018) models the choice of production rate Ft, condi-
tional on some initial stock of drilled wells with production capacity K0, as the problem of maximizing∫∞0 U(Ft)e
−rtdt, subject to the constraints Ft ∈ [0,Kt] and Kt = −λFt, where λ denotes the exponential
production decline rate whenever Ft = Kt. Letting θt denote the current shadow value of capacity Kt, the
FOC governing the optimal production rate is then that the capacity constraint binds when U ′(Ft) > λθt.
That is, it will be optimal for firms to produce at full capacity if U ′(Ft)—the oil price—exceeds the value
of the capacity that is reduced in the course of production (producing at a rate Ft reduces capacity by λFt,
and that capacity is valued at θt dollars per unit).
A testable implication of this production model is that, during the 1990–2007 time period studied in
Anderson et al. (2018), the U ′(Ft) > λθt condition should hold, since firms appear to have set Ft = Kt
throughout this time. To conduct this test, the paper estimates λ = 0.1 based on the observed production
decline rate and estimates θt using a second implication of the model: the value θt of a marginal unit of
capacity should equal the expected future stream of oil prices, discounted at λ + r. Anderson et al. (2018)
then shows empirically that, under a variety of reasonable assumptions on the appropriate discount rate, the
U ′(Ft) > λθt condition holds throughout the sample.
The upshot is then that, barring extreme periods in which the price of oil is forecast to increase rapidly
for a long period of time, the problem of optimizing oil production from previously-drilled wells is actually
rather boring. Price-taking firms should produce wells at their geologic capacity constraint, and production5For an accessible primer on petroleum geology and decline curves, see Hyne (2001).
13
will then naturally decline over time. And micro-data indicate that this is exactly what U.S. oil producers
do. Moreover, this empirical result has been replicated for modern shale oil and gas wells by Newell and
Prest (2019) and Newell et al. (2019).
Drilling decisions, in contrast, are where the action is and where the dynamic Hotelling-style incentives
lie. In Anderson et al. (2018), firms drill new wells to increase their production capacity, facing a marginal
drilling cost that is strictly increasing in the rate of drilling, and facing a finite stock of wells that they can
drill. Denoting the drilling rate by at, the initial capacity of a new well by X , and marginal drilling cost
by d(at), the return to drilling a marginal well is given by θtX − d(at). Anderson et al. (2018) shows
that, because the stock of wells to drill is a scarce resource, this marginal value must increase over time at
the discount rate r. Thus, the Hotelling logic applies to drilling timing decisions, rather than to production
decisions once drilling has taken place.
While the model in Anderson et al. (2018) is deterministic, the paper closes by simulating drilling,
production, and price outcomes when the equilibrium path is subjected to one-time, unanticipated demand
shocks. The simulated responses to these shocks can result in price paths that, if anticipated following the
shocks’ impact, would naturally lead to switches between backwardation and contango in futures markets.
Namely, following a positive demand shock, the oil price must increase on impact because production
cannot. Firms will then respond by increasing their rate of drilling, consistent with observed drilling data.
This increased drilling will eventually increase the production rate, causing the price of oil to gradually
fall after the initial impact of the shock. If agents anticipate this behavior, oil futures markets should be
backwardated immediately following the shock, which is consistent with the backwardation observed in
historical futures price data.
Anderson et al. (2018) therefore makes two high-level contributions. First, it shows that a Hotelling-style
model actually can provide useful empirical predictions for the oil and gas extraction industry, provided that
the model incorporates the industry’s essential institutional features and constraints. Second, it provides a
theoretical and empirical foundation for why economists studying the oil and gas industry should focus their
attention on drilling investments rather than on production decisions once wells have been drilled. Many
of the papers we discuss below that examine mineral leasing policies, productivity, and firms’ strategic
behavior adopt this approach.
2.1.3 Real options: drilling in the presence of stochastic oil prices
Crude oil markets are of course stochastic, rather than deterministic as modeled in Anderson et al. (2018).
Because drilling an oil well is a classic example of an irreversible investment—there is no way to “un-drill” a
well and recover the investment cost—firms holding drillable acreage should account for this volatility when
deciding when to drill their wells. Specifically, real options theory (Dixit and Pindyck, 1994) highlights that
high anticipated future oil price volatility should, all else equal, increase the value of holding investment
for the future, making drilling today less attractive. The intuition for this result does not come from risk
aversion, but rather the fact that high volatility increases the potential upside from sinking the investment in
the future, while the downside of waiting is bounded by the fact that the firm can always choose not to drill
if the oil price turns out to be low (achieving profits of zero rather than strictly negative profits).
14
Kellogg (2014) examines whether the drilling decisions of oil and gas companies actually comport with
the prescriptions of real options theory. The paper aims to test not only whether firms reduce their rate of
drilling when expected oil price volatility is high, but whether the magnitude of this effect matches what is
implied by dynamic optimization. The paper ultimately answers this question in the affirmative, helping to
underscore the usefulness of dynamic models for explaining firms’ investment behavior.
To execute its empirical test, Kellogg (2014) takes advantage of two helpful features of the oil and
gas setting. First, the presence of both oil futures markets and oil futures options markets mean that it
is possible to obtain independent measures of the market’s expected future oil price and expected future
price volatility (the latter of which can be inverted out from options prices, using a variant of the Black and
Scholes (1973) financial options pricing model). Second, the paper takes advantage of the fact that U.S. oil
producers behave competitively in the output market, so that firms’ investment problem can be modeled as
a single-agent dynamic program rather than as a more challenging dynamic game.6
In a simplified version of the model employed in Kellogg (2014), the optimal drilling program should
be given by the solution to the following Bellman equation:
V (P, σ) = max{π(P ), δE[V (P ′, σ′)]}, (5)
where P and σ are the state variables denoting the current price of oil and the firm’s belief about oil price
volatility, respectively. π(P ) denotes the current-period payoff to drilling, and δ is the discount factor. The
oil price P directly affects the profits from drilling, but volatility σ does not. Instead, σ affects the variance
of the distribution of oil prices that the firm believes it will face next period.
Kellogg (2014) then tests whether firms’ beliefs about oil price volatility, as implied by their drilling
behavior, line up with actual oil price volatility data from the futures options market. The paper models
firms’ volatility belief at a given time t as a linear combination of the sample average logged volatility, log σ
and the difference log σdt between the actual volatility at t and this average. That is:
log σt = log σ + β log σdt . (6)
If β = 1, then firms’ beliefs (as implied by their drilling behavior) align with the market. In contrast,
the value β = 0 would imply that firms behave as if oil price volatility does not vary over time.
Kellogg (2014) estimates β using monthly data on drilling, oil prices, and implied price volatility from
1993–2003. Both the price and price volatility vary substantially during this time period, with volatility
experiencing notable increases during the 1998–1999 oil price crash and following the 9/11 attacks. To
enable the model to rationalize the observed drilling data, the final empirical model also includes a time-
varying unobservable that represents firms’ beliefs about the quantity of oil that will be produced from each
well.
Kellogg (2014) finds that firms respond to volatility shocks in a way that closely matches the theory: the
estimated value of β is 1.12. The paper then surmises that firms may behave this way because there is sub-6To further ensure that a single-agent model is appropriate, Kellogg (2014) limits its sample to established Texas oil fields
that are operated by a single firm, thereby avoiding issues of information externalities (we discuss papers tackling these issues insubsection 2.3).
15
stantial value at stake in getting drilling decisions right. For instance, the paper constructs an example using
in-sample variation in volatility, showing that recognizing that volatility is relatively high—and therefore
delaying drilling—can increase a drilling opportunity’s value by 27%.
2.1.4 OPEC, market power, and drilling sequencing
The model in Anderson et al. (2018) assumes that oil and gas firms act as price-takers. This assumption
is reasonable for the many privately-owned firms that produce oil and gas, but it is likely problematic for
the large oil-producing nations with nationally-owned oil companies that participate in the Organization of
Petroleum Exporting Countries (OPEC) cartel. The market power possessed by this large cartel, or even
unilaterally by a very large oil producer like Saudi Arabia, might plausibly lead to distortions in crude oil
extraction.
In a competitive market, low-cost resources should be produced before high-cost resources, as shown by
Herfindahl (1967) using a Hotelling-style model in which different deposits produce perfectly substitutable
products but at different marginal costs. But if the low-cost producers possess market power, they may find
it profitable to delay extraction, leading to high-cost resources being produced simultaneously with or even
after low-cost resources. This out-of-merit production ordering is a form of allocative inefficiency.
Asker et al. (2019) attempts to quantify the extent of this inefficiency using data from the global oil
market from 1970 to 2014, during which time OPEC held a roughly 40% share of world oil production.
Because well-level data are not available at a global scale, Asker et al. (2019) uses proprietary oil field-
level data on production and extraction costs from Rystad Energy, an energy consultancy. A necessary
limitation of these data is that it is not possible to separately model the costs of new drilling investment
versus production from existing wells and facilities.
Asker et al. (2019) uses the Rystad cost data to construct a counterfactual, surplus-maximizing global oil
extraction plan in which extraction proceeds in strict order from the lowest-cost to highest-cost fields. The
key assumption the paper makes to enable this approach is that marginal costs are constant within-field (else,
it may be optimal to produce from multiple fields simultaneously). The paper can then quantify the extent
of mis-allocation by comparing costs under the observed extraction order to costs under this counterfactual,
holding constant the total amount of extraction each year.
Because the Rystad data indicate that production costs in many of the major OPEC countries are sub-
stantially lower than costs outside of OPEC, Asker et al. (2019) finds that the costs of OPEC-driven mis-
allocation are substantial. Asker et al. (2019) values OPEC-driven mis-allocation at $105 to $163 billion
during 1970–2014.7 The paper therefore highlights how OPEC’s exercise of market power can lead to
large allocative efficiency losses by distorting extraction timing relative to that implied by a Hotelling-style
benchmark model in which firms behave competitively. Note that because the paper’s counterfactuals hold
total global extraction constant, this welfare loss is in addition to the standard welfare loss from market
power associated with any decrease in the total volume of oil produced each year (though such a loss may
be substantially offset by the environmental damages associated with oil production and consumption).7These values are deflated to 2014 dollars.
16
Asker et al. (2019)’s estimated OPEC-driven distortion omits distortions related to mis-allocation within
OPEC itself as well as distortions within the set of non-OPEC countries. The paper’s estimate of the overall
surplus distortion that includes these other forms of mis-allocation (including within-country mis-allocation)
is $744 billion. However, some of this figure may be due to measurement error in the Rystad data or to
expectational errors that would occur even in the absence of any market failure.8 Nonetheless, Asker et
al. (2019) implies that a variety of frictions—including but not limited to market power—may impede oil
and gas development from proceeding in a fully efficient manner. Some of the papers we discuss below
build on this result by pointing to specific frictions that inhibit efficient lease allocation or rapid productivity
improvements, thereby diminishing the sector’s overall efficiency.
Finally, we note that the results in Asker et al. (2019) also have rather negative implications for climate
policies aimed at reducing global oil consumption. The OPEC-driven mis-allocation in Asker et al. (2019)
arises from the result that, for many large OPEC nations, the marginal cost of producing additional crude oil
is substantially less than recent oil prices. This result will make it harder for alternative fuels to effectively
compete with oil, since displacing oil at scale will ultimately require out-competing oil at cost, not at current
prices. Quantifying the climate policy implications of the gap between oil prices and the low marginal
extraction costs in OPEC nations therefore strikes us as a valuable contribution for future research.
2.2 Oil and gas mineral leasing
The owners of underground mineral resources—whether they be sovereign governments, state governments,
or private individuals—typically do not themselves possess the necessary expertise or capital required to
carry out extraction in a cost-effective manner (or at all).9 To realize value from their resources, mineral
owners instead lease them to private extraction firms. These leases play a central role in determining when
particular resources are developed, which firm develops them, and how resource rents are shared.
U.S. federal oil and gas leases—and in particular lease auctions in the Outer Continental Shelf (OCS)
of the Gulf of Mexico—have historically received a great deal of attention from IO economists. An earlier
volume of this handbook (Hendricks and Porter, 2007), as well reviews in Porter (1995) and Haile et al.
(2010), summarizes the catalog of research on bonus bidding in U.S. federal OCS auctions through 2010.
Three important lessons from this body of work are that: (1) exploratory (“wildcat”) lease sales in the
OCS exhibited large bid dispersion, substantial competition, and the government capturing a large share of
the tracts’ value; (2) sales of “drainage” tracts located adjacent to producing tracts were less competitive,
since the neighboring lessees could use their information advantage to capture rents; and (3) firms submit
bids consistent with their recognition of the “winners’ curse” in common value settings. We refer readers
interested in this literature to the excellent reviews highlighted above.
In this sub-section, we will discuss recent work in industrial organization that extends the literature
above by expanding our understanding of bidding firms’ valuations and informational environment, and of8Expectational errors are out-of-order extraction that occurs because ex-post costs turned out to differ from costs that extractors
expected to incur prior to investing in a field.9Countries with national oil companies are an exception. However, because little data are available regarding the operations of
such companies, the IO literature has instead overwhelmingly focused on settings—especially the United States—where govern-ments or private owners lease their interests to extraction firms.
17
aspects of oil and gas leasing beyond the up-front bonus payments. This research has leveraged new datasets
from states’ oil and gas lease auctions, which exhibit experimentation in auction formats that is largely
absent from federal OCS auctions, and from private oil and gas leasing, which takes place via decentralized
negotiations rather than formal auctions.
2.2.1 Royalties and primary terms
We begin by discussing recent work oil and gas leases’ royalty clauses, which dictate that the lessee firm
must pay a royalty share of its oil and gas revenue to the mineral owner. While royalty clauses often allow for
some deductions related to transportation and processing costs, they do not allow the firm to take deductions
for the major, up-front costs of drilling and completing wells.10 Thus, the royalty effectively acts as a tax on
revenue, distorting the firm’s incentives relative to the case where it is the residual claimant on all drilling
and extraction activity. In particular, the royalty will induce the firm to delay drilling (or perhaps not drill
at all) and may potentially also induce the firm to reduce other inputs—such as the intensity of a hydraulic
fracturing treatment if one is being applied—conditional on drilling and completing a well.
So why then do leases include royalties at all? The answer is provided by theory developed in Laffont
and Tirole (1986), Riley (1988), and Hendricks et al. (1993). If there are only a handful of firms who are
interested in leasing the parcel, and if those firms have differing valuations for the parcel, then selling the
lease for only the bonus bid, without a royalty, will necessarily leave the winning firm with information
rents. One way to reduce these information rents is to link the payments made by the winning firm to the
realized value from the tract (Milgrom and Weber, 1982). The royalty clause in an oil and gas lease does
precisely this. However, including the royalty comes at the cost of distorting the firm’s extraction effort.
These effects can be illustrated in a simple, stylized model. Suppose there is only one potential lessee
firm, and the mineral owner can make the lessee a take-it-or-leave-it offer that is a combination of an upfront
payment p and royalty r. The lessee’s profits from the lease, should it accept the offer, are determined by:
• v: the firm’s private value on the reserves. The mineral owner believes v is distributed F (v), with
support on [vL, vH ], where vL > 0.
• E: the extraction “effort” undertaken by the firm. In practice, effort will take the form of drilling the
well sooner rather than later, or increasing the intensity of hydraulic fracturing. The empirical papers
that we discuss below will model these forms of effort explicitly, but for now we simply consider
effort as measured by the scalar E ≥ 0 and abstract away from any dynamics.
The firm’s profits π are then given by
π(E|v, p, r) = (1− r)vEα − CE − p, (7)
where α ∈ (0, 1) and C > 0. Conditional on accepting the lease, and on r < 1, the optimal effort E∗ is then
10Allowing cost deductions—effectively making royalties more like profit taxes—would improve the efficiency of these con-tracts. In practice, however, accurate cost reporting can be difficult to monitor, audit, and enforce, especially when the lessee firmoperates leases across many different owners, and when the owners are private individuals with limited expertise and resources.
18
strictly greater than zero, and it is a strictly decreasing function of the royalty r. The firm will then accept
the lease offer if π(E∗|v, p, r) ≥ 0.
Let v(p, r) denote the lowest participating firm type given the bonus p and royalty r. The expected value
Vo of the lease to the mineral owner is given by:
Vo(p, r) =
∫ vH
v(p,r)(p+ rvEα)dv (8)
To see that the optimal royalty is non-zero, we can differentiate equation (8) with respect to r and
evaluate it at r = 0. Doing so yields:
dVo(p, 0)
dr=
∫ vH
v(p,0)vEαdv, (9)
which is strictly greater than zero. From the mineral owner’s perspective, the revenue gains from increasing
the royalty rate (starting from 0%), and thereby reducing the firm’s information rents, are first-order. The
cost of the induced effort distortion, however, is second-order. Thus, the revenue-maximizing royalty rate
for the owner will generally lie strictly between 0% and 100%.
Two questions then follow from this intuition. First, in typical oil and gas lease auctions that fix the
royalty and ask firms to submit bonus bids, what is the revenue-maximizing royalty rate to set? And second,
what are the benefits or drawbacks of alternative auction formats that ask bidders to submit royalty bids
(either with a fixed bonus or with a bonus that is also a bid variable)? We next discuss a set of recent papers
that have been the first to address these questions empirically.
Optimal royalties in fixed-royalty bonus bid auctionsIn the commonly-used fixed-royalty bonus bid auction format, the revenue-maximizing royalty rate
depends on the amount of private information possessed by firms (which is competed away as the number
of bidding firms increases) and the extent to which the royalty distorts drilling activity after the lease is
awarded. Bhattacharya et al. (2020) and Ordin (2019) address this question using data from lease auctions
and drilling activity on state-owned mineral parcels in New Mexico. The workhorse of both papers is a
structural model that links: (1) a lease auction in which firms with heterogeneous valuations bid on tracts,
taking the royalty rate as given; and (2) a model of the winning firm’s drilling timing problem. The papers’
goal is to use this combined model to simulate counterfactuals with different royalty rates and assess how
royalties affect bidding, the state’s revenues, and drilling activity.
In Bhattacharya et al. (2020)’s and Ordin (2019)’s model, the firm’s choice of when to drill (or to not
drill at all) is the analog to the effort E in the simple model presented above. This drilling timing problem
is represented in the model as a single-agent, finite-horizon dynamic optimization problem, via a Bellman
equation. A simplified version of this problem is given by:
V (p, q, c, r) = max{(1− r)pq − c, δE[V (p′, q, c, r)]}, (10)
where p denotes the oil price (which is stochastic), q denotes the firm’s belief about quantity, c is the drilling
19
cost, and r is the royalty rate. The solution to this problem not only gives drilling probabilities each period as
a function of output prices and lease terms, but also firms’ valuation of each tract at the time of the auction.
The model then uses these valuations in the first-stage auction game. This auction is modeled a a standard
common value auction, given the solution to the Bellman equation (10) and a parameter governing firms’
uncertainty about the tract’s quantity q at the time of bidding.
Bhattacharya et al. (2020)’s and Ordin (2019)’s estimation and simulation of the combined auction
and drilling timing model are themselves a methodological innovation, since to the best of our knowledge
previous work has not combined these models into a single package. This approach is likely to be useful
not just for auctions of natural resource extraction rights, but also for other settings in which a principal
grants an agent the right to develop or execute a project, the realized value is ex-post contractible, and the
agent has some discretion as to the quantity or quality of the work. Both papers use data on firms’ bonus
bids (New Mexico uses sealed-bid auctions and makes information on winning and losing bids available),
drilling timing, oil production, oil prices, and royalties (which vary across tracts) to jointly estimate the
parameters of the model in a single step by matching simulated moments to the data.11
Unlike many auction papers, the distribution of bidders’ values at the time of the auction is not itself
a primitive to be estimated, since these values are the solution to the Bellman equation (10). Instead, the
papers must estimate the distribution of actual tract quantities, along with the noise in bidders’ signals of
tract quantities and firms’ drilling costs. Obtaining the distribution of quantities conditional on drilling
is straightforward because this object is observed directly in the data. The unconditional distribution is
identified using information on firms’ drilling timing decisions (or decision not to drill at all), where the oil
price p is used as an excluded variable: it is a determinant of drilling time that does not affect output.
The estimation method used in Bhattacharya et al. (2020) proceeds roughly as follows. Given a set of
parameters, the model can be simulated via backwards induction. That is, the paper first solves the drilling
timing problem given by equation (10) for a range of quantities q. Then, given a candidate distribution of q,
the paper repeatedly draws firms’ signals of q and simulates common-value auctions (and then the drilling
decisions that follow). Estimation of the parameters and the distribution of q then proceeds by matching
simulated moments to actual moments in the data. These moments include the highest, second-highest, and
average bid; moments of the quantity distribution; moments for drilling delays; and interactions.
Both Bhattacharya et al. (2020) and Ordin (2019) use the estimated model to find that the revenue-
maximizing royalty rate for the state of New Mexico is 29%, which is modestly higher than the royalty
rate the state uses in practice. Ordin (2019) emphasizes that the revenue-generating property of the royalty
comes at the cost of significantly reducing drilling: it estimates that eliminating the royalty entirely would
increase the unconditional probability that a tract is drilled from 0.096 to 0.154 (a 60% increase).
Taken together, these papers therefore provide quantitative guidance for how changing royalty policies
will influence the government’s revenues and drilling activity. Providing results of this nature is not possible
in the absence of a model that links firms’ behavior in the leasing auction to their ex-post drilling actions,
and these papers are the first to provide an econometric approach to doing so.11Following other work using optimal stopping problems to model drilling, such as Kellogg (2014), Bhattacharya et al. (2020)
and Ordin (2019) estimate the oil price transition process in a preliminary step.
20
Auctions with royalty biddingAnother advantage of the model that Bhattacharya et al. (2020) and Ordin (2019) develop is that they can
use it to simulate counterfactual policies other than the standard bonus bid auctions that New Mexico (along
with other states and the U.S. federal government) uses. These simulations are motivated by theoretical work
in DeMarzo et al. (2005) and Skrzypacz (2013) on securities auctions that link the bidders’ total payments to
cash flows that are realized in the future. One alternative security auction form that could be used in oil and
gas leasing is an “equity” auction, in which firms would bid in a royalty rate rather than a bonus. DeMarzo
et al. (2005) shows that in the absence of moral hazard, the equity auction is likely to maximize the seller’s
revenue. Skrzypacz (2013), however, emphasizes that equity auctions can decrease both the seller’s revenue
and total surplus if the high rate of revenue sharing significantly distorts the agent’s actions after the auction.
Bhattacharya et al. (2020) and Ordin (2019) are two of the first papers to quantify the impacts of these
alternative auction forms using a model estimated from real-world data. Both papers find that equity auctions
would perform terribly in New Mexico oil and gas leasing. In the simulated counterfactuals, firms bid
high royalty rates that can exceed 40%, ultimately resulting in little drilling taking place and therefore little
revenue flowing to the government. Kong et al. (2020) finds similar results using data from state-run auctions
in Louisiana. There, firms bid in both a bonus and a royalty rate, and Kong et al. (2020) finds that a simpler
bonus bidding mechanism with a fixed royalty would increase the state’s revenue, the likelihood that tracts
are allocated to the firm with the highest valuation, and the likelihood that tracts are drilled (Kong et al.
(2020)’s model involves private rather than common valuations).
Primary termsIn addition to bonus bids and royalties, oil and gas leases also include primary terms that limit the number
of years the lessee has to drill and complete at least one productive well. If the firm does not drill by the end
of the primary term, it must relinquish the lease, and the mineral owner is then free, if desired, to re-lease
the parcel (either to the same firm or to a different one). This deadline can substantially alter firms’ drilling
incentives. For instance, in Bhattacharya et al. (2020)’s and Ordin (2019)’s simulations, firms’ likelihood of
drilling, conditional on not yet having drilled, increases sharply as the deadline approaches.
Herrnstadt et al. (2020) studies the impacts of primary terms in the context of private natural gas leasing
in the Haynesville Shale in northwest Louisiana. One feature of this environment is that private parcels are
typically too small on their own to support drilling. The state therefore prescribes a process by which leases
may be force-pooled into square-mile drilling units, and that drilling a productive well anywhere in the unit
holds all the leases in the unit. Importantly, the operating firm may then drill subsequent, follow-up wells
elsewhere in the unit. The incentive to preserve the option for future drilling is strong, and Herrnstadt et al.
(2020) begins by showing that in Haynesville drilling units there is clear bunching of drilling timing into the
months just prior to the first lease expiration date in the unit.
While the bunching of drilling timing may seem distortionary ex-post, Herrnstadt et al. (2020) next illus-
trates why primary terms may actually enhance both efficiency and the owner’s revenues ex-ante. Starting
with an analytical model that builds on Laffont and Tirole (1986) and Board (2007), Herrnstadt et al. (2020)
shows that if the owner induces the firm to accelerate drilling by including a drilling subsidy in the lease
21
contract, expected revenue can increase because the subsidy counteracts the delay incentive generated by the
royalty. Although an explicit subsidy may often be impractical due to liquidity constraints or other factors,
primary terms can fulfill the same objective, albeit in a coarse way (similar to “notched” tax policies in the
public finance literature (Kleven, 2016)). This intuition, while most directly applicable to oil and gas leases,
may also be relevant to other settings, such as master franchise contracts or licenses to adapt creative works,
in which principals sell time-limited development options to agents.
Herrnstadt et al. (2020) then develops and estimates a structural model to explore how revenues, total
surplus, and drilling are affected by primary terms. The core of the model, like in Bhattacharya et al. (2020)
and Ordin (2019), is the firm’s optimal stopping problem, which has a finite horizon when there is a primary
term or an infinite horizon if not. Herrnstadt et al. (2020) adds to this framework the feature that, in the
event the primary term expires, the owner and firm can sign a new lease, requiring another bonus payment.
Consistent with the paper’s analytical results, Herrnstadt et al. (2020) finds that primary terms can in-
crease the owner’s expected revenue by 1%–3%, relative to a contract with just a royalty and no primary
term deadline. Total surplus also increases because drilling is pulled forward in time, counteracting the
delay induced by the royalty.
Herrnstadt et al. (2020) also finds, however, that primary terms are likely to reduce the owner’s revenue
when the first well drilled in a unit creates an indefinite option to drill additional wells in the future. In
that situation, owners are better off if they can impose primary terms on all potential wells that might be
drilled, not just the first one. Regulations in Louisiana proscribe such contracts, but these results relate to the
increasing use by other states’ private mineral owners of “retained acreage” clauses that allow firms to hold
only lease acreage that is proximate to a drilled well, and not acreage that might be used for future wells.
2.2.2 Information, common value auctions, and royalties
The leasing of wildcat (exploratory) tracts is a classic common-values auction setting. In a common-values
environment, the bids of losing firms help inform the winning firm’s beliefs about how productive the tract
will be. These beliefs in turn affect the winning firm’s likelihood of drilling. An important question for the
mineral owner is then whether to release the values of losing bids to the winner. Doing so might increase
the winner’s probability of drilling—and therefore also the owner’s royalty revenue—if the losing bids are
relatively high (i.e. not too far below the winning bid), but they may decrease the probability of drilling if
they are relatively low. Moreover, the owner’s policy on bid data release might plausibly affect firms’ bidding
strategies in the auction. Which of these effects dominates is an empirical question, and this question is, to
the best of our knowledge, addressed for the first time in Nguyen (2021).
Nguyen (2021) studies U.S. federal OCS leasing from 2000–2019, during which time only 24.5% of
leased tracts were actually drilled. Throughout this period, the government’s policy was to release the values
of losing bids, and the paper begins by showing that the likelihood of drilling is increasing in the value of
the highest losing bid (conditional on the winning bid), suggesting a role for information transmission. The
paper’s goal is to evaluate what drilling and government revenues would have been in a counterfactual in
which the government did not disclose the losing bids.
As is standard in models of common value auctions, each bidding firm i in Nguyen (2021) is modeled
22
as receiving a signal Si of the tract’s true expected oil and gas revenue Q. If the winning firm drills, it
must pay a drilling cost C that is drawn after the auction from a distribution FC that is independent of
Q. This timing and independence assumption is important to the tractability of the model (and ultimately
identification) because it means that firms receive no additional information about Q after the auction, and
that there is no private value component to firms’ valuations of the tract. The model further simplifies the
winning firm’s drilling problem by treating it as a static binary choice between not drilling at all versus
drilling (and realizing profits equal to (1− r)Q− C, where r is the royalty rate).
The key object that must be identified in order to conduct the counterfactual analysis is the function
governing firms’ posterior beliefs about Q, conditional on the private signals Si, i.e. V (s1, ..., sN ) =
E[Q|S1 = s1, ..., SN = sn]. However, the fact that Q is not observed for most tracts (since they are not
drilled) poses an identification challenge. To identify the distribution ofQ, Nguyen (2021) relies on variation
in drilling costs as measured by the rental cost for drilling rigs, arguing that these costs affect drilling timing
but not Q.
With its model estimated, Nguyen (2021) then examines how outcomes would have differed had the
government not disclosed the losing bids. The main result from this counterfactual is that bonus bid revenue
modestly declines (because firms recognize that they will be more likely to make “mistakes” when making
drilling decisions), but drilling and royalty revenue increase substantially, so that the government’s revenue
per deepwater tract increases by $734,000 on average. The drilling rate increases because disclosure of
losing bids is more likely, on average, to stop a winning firm from drilling than to induce a winning firm to
drill. This asymmetry is especially strong on more productive tracts, leading to the substantial increase in
production from adopting a non-disclosure policy.
2.2.3 Uncertainty about the number of bidders
Government-run auctions of oil and gas leases sometimes attract many bidders, sometimes attract just one
bidder, and sometimes attract none at all. In the commonly-used sealed bid auction format, the number of
bidding firms is likely to be not precisely known by the bidding firms. How might this uncertainty affect
firms’ bidding behavior, particularly when firms are risk-averse?
Kong (2020) addresses this question by studying oil lease auctions for state-owned land in New Mexico,
from 2005–2014. The paper begins by documenting that there are many sealed-bid auctions in the data in
which there was only one bid, yet that bid was several times greater than the publicly-announced reserve
price. A firm that knew it was the only bidder would never bid this way (since it could win the tract by
simply bidding the reserve), so these auctions are evidence of firms’ uncertainty regarding the number of
competitors they face.
Kong (2020) then develops a private-value model with non-selective entry, in which bidders are permit-
ted to be risk-averse.12 There are N potential bidders, which have values v distributed F (v). Each potential
bidder enters the auction with probability p and has a monotonic bidding strategy b(v; p). In this model, if
a bidder is risk-averse, it may be optimal to place a bid b that is substantially higher than the reserve, even12The paper also allows for asymmetric valuations and entry probabilities for subgroups of bidders. We simplify the discussion
by abstracting away from this feature of the paper.
23
if the expected number of other entrants is not much greater than zero. Placing such a high bid effectively
insures the firm against the risk of not winning the parcel. Importantly, this effect exists in first-price sealed
bid auctions but not in an English ascending-bid or second-price auction, in which firms’ weakly-dominant
strategy is to bid their value v if they enter.
To estimate the model, the Kong (2020) first estimates entry probabilities as a function of tract charac-
teristics, using information on the number of potential bidders versus the number of actual bidders in each
auction. To separately identify the distribution of firms’ valuations from firms’ risk aversion, the paper then
takes advantage of the fact that New Mexico uses both English auctions and sealed-bid auctions. Under an
assumption that the selection of auction format is as good-as-random, conditional on tract observables, the
open-outcry auctions pin down the distribution of valuations (Athey and Haile, 2002), and then the sealed
bid auction data identify the degree of risk aversion.
Using the estimated model, Kong (2020) shows that the combination of risk aversion with uncertainty
about the number of bidders leads to the sealed-bid format generating substantially larger revenues than the
open-outcry format, especially in situations where the expected number of bidders is low. Overall, Kong
(2020) estimates that converting the sealed-bid auctions in the data to an open-outcry format would decrease
the average bonus bid by nearly $27,000 per tract, equal to 21% of the average winning sealed-bid. The
impact of the format change is often on par with that obtained by reducing the number of bidders by one.
The large magnitudes of these results indicate a substantial revenue advantage of sealed-bid auction formats,
relative to English auctions, in settings with little competition and uncertain bidder entry.
2.2.4 Sequential lease auctions and firms’ valuations of neighboring tracts
Governments will sometimes auction neighboring oil and gas tracts in a sequential manner. That is, after a
tract is auctioned in a lease sale, a the government may auction a neighboring tract in a following sale. Firms’
valuations of the two tracts are likely to be related for at least two reasons. First, because the underlying
geology is spatially correlated, firms’ valuations for the two tracts are likely to be affiliated (i.e., each firm’s
value for one tract will be correlated with its value for the other). Second, because there may be synergies
(e.g., economies of scale) from developing neighboring tracts together, the winner of the first tract may
increase its valuation of the second tract. Is auctioning such tracts sequentially the revenue-maximizing
way to go for governments, or should they instead auction the tracts together as a bundle? The answer to
this question is likely to be relevant to settings other than oil and gas leases. For instance, FTC spectrum
auctions feature licenses of spectrum for geographically adjacent areas, for which bidders’ values are likely
to be correlated due to affiliation, synergies, or both.
Kong (forthcoming) studies this question using data from New Mexico auctions for state-owned tracts.
New Mexico often adopts the practice of offering “pairs” of neighboring tracts on the same day, with one
tract being offered first in a sealed-bid auction and the other subsequently offered in an English auction.
Thus, firms will know whether or not they won the first tract before bidding on the second.
The core empirical challenge in Kong (forthcoming) is to separately identify affiliations in values from
synergies across the tracts. This challenge is akin to the classic problem discussed in Heckman (1981)
of separating persistent heterogeneity from state dependence in panel choice data. Kong (forthcoming)
24
addresses this problem via a regression discontinuity design. The paper defines, for each firm in the first
auction of each pair, a running variable z that is the difference between the firm’s bid and the highest bid
among the other firms. The paper then shows that the probability the firm wins the second auction, as a
function of z, is discontinuous at z = 0, consistent with significant cross-track synergies.
To assess the counterfactual of a bundled auction, Kong (forthcoming) then nests its regression disconti-
nuity design within a private values model of bidding behavior, wherein bidders in the first auction anticipate
that winning will yield benefits in the second auction. The estimated model indicates that both synergies and
affiliated values are important in this setting, though the latter are relatively more important. Kong (forth-
coming) then simulates counterfactuals to show that auctioning tracts as a bundle would increase the state’s
revenue by 7% on average. This result is driven by bundling’s guarantee of achieving synergies, as well as
the traditional effect of bundling that it reduces overall dispersion in values (so long as bidders’ values for
the two tracts in each pair are not too strongly affiliated, see for instance Adams and Yellen (1976)). On the
other hand, bundling decreases the total surplus from the tracts, by 2 to 3%, since the allocative efficiency
loss from being able to separately allocate each tract to the highest-value firm outweighs the gain from
locking in synergies. Thus, Kong (forthcoming) highlights that the optimal bundling strategy for a seller
in a multi-unit auction may depend on whether the seller’s goal is revenue-maximization or total surplus
maximization.
2.2.5 Auctions vs unstructured oil and gas leasing
The vast majority of empirical IO papers on oil and gas leasing study organized government-run auctions
of tracts (Herrnstadt et al. (2020) discussed above is one exception). Such settings are ripe for study be-
cause the auction rules are publicly announced, bonus bid data can be obtained by researchers, and a single
government office can typically provide data on a large number of auctions.
Nonetheless, most oil and gas leases in the United States are for privately, not publicly owned mineral
resources. These leases overwhelmingly do not transact in organized auctions but are instead sold in a de-
centralized marketplace via negotiations. Typically, negotiations for an oil and gas lease of private minerals
are initiated by the firm rather than the mineral owner, and survey evidence suggests that many mineral
owners execute a lease contract with the first firm that knocks on their door (Ward and Kelsey, 2011).
While leases for privately-owned minerals share the same basic bonus + royalty + primary term struc-
ture as do leases for publicly-owned minerals, the markedly different way in which these leases are allocated
might plausibly lead to substantially different outcomes for the owner’s revenue and resource development.
Yet evaluating differences in outcomes between these two formats is challenging. Direct comparisons be-
tween private and public parcels are subject to confounds from omitted variables that might be correlated
with mineral ownership. And simulating counterfactuals is burdened by the problem that there is a dearth of
applicable models of bilateral negotiation of contingent contracts when the buyer has private information.
Covert and Sweeney (2019) makes progress on this question by taking advantage of an unusual situation
in Texas whereby some parcels were plausibly exogenously assigned to an auction format, while others were
not. For reasons dating back to Texas’s time as a Spanish colony, its status as sovereign nation in the 19th
century, and the Texas Relinquishment Act of 1919, some of Texas’s state-owned minerals exist on split
25
estates where private individuals own the surface rights. On some of this land the surface owners negotiate
mineral leases on their own in exchange for half of all proceeds, while on the remainder the state retains a
full 100% interest and allocates leases via organized, fixed-royalty bonus bid auctions.
Covert and Sweeney (2019) argues that the assignment of split estate parcels to each of these lease
allocation formats is plausibly exogenous with respect to other determinants of the value of shale oil and
gas deposits (which have only recently become exploitable). Then, because the bonus bids for the privately
negotiated leases are publicly recorded (since the state has a 50% interest in the revenues that flow from
them), Covert and Sweeney (2019) can, unusually, make a direct comparison between the revenues from
privately negotiated versus publicly auctioned leases. The paper finds that bonuses on auctioned leases are
dramatically larger: $584 to $1,009 more per acre on average. These values are equivalent to 55%–95% of
the average per-acre bonus on private leases. Auctioned leases also have, on average, slightly higher royalty
rates, so these bonus differences actually understate the revenue effects of organized auctions.
Covert and Sweeney (2019) also finds that auctioned leases are 8 to 18 percentage points more likely
to be drilled and increase oil and gas production by 44 to 74% of the average output of negotiated leases.
These effects are consistent with auctions being able to better select firms that will be productive lessees.
Covert and Sweeney (2019) hypothesizes that these results reflect the decentralization and high search
costs of private oil and gas leasing markets, where firms must undertake effort to identify leasable parcels
and perform title searches. Thus, this market is potentially characterized by “non-sequential” search in
which more productive firms are not necessarily more likely to be the first to contact the lessor. The up-
shot is then that there may be large gains—for both lessors’ revenues and total surplus—from policies that
either encourage or mandate a more centralized lease allocation mechanism for private oil and gas mineral
interests.
2.2.6 Reassignment of leases after the initial sale
One question raised by Covert and Sweeney (2019)’s results on allocative inefficiencies in the private leasing
markets is why, given the potentially large productivity gains, the initial lessees do not sell their leases to
more productive firms. Such sales—termed “lease reassignments” in the industry—are generally permitted
by both private and public oil and gas leases, but the results in Covert and Sweeney (2019) suggest that
reassignments are not occurring a frictionless manner.
Progress on understanding the reassignment process has been made by Brehm and Lewis (forthcoming),
which takes advantage of variation generated by 1970s lease allocations of federal land in Wyoming that
used lotteries rather than auctions. Firms and individuals were able to enter these lotteries for the low fee
of just $10, which naturally attracted a large number of bidders. As a consequence, the lotteries were often
won by individuals who were completely unqualified to actually develop the parcel.
The dataset used by Brehm and Lewis (forthcoming) includes information on each tract’s lottery winner
as well as the second and third-place winners (whose lots were drawn in case the first-place winner did not
pay the fee). The paper’s empirical strategy is then to regress outcomes (lease reassignment, drilling, and
production) on a treatment indicator for whether the winning bidder was an genuine oil and gas firm rather
than an unqualified individual. To address potential selection problems associated with endogenous entry
26
into the lotteries, the main specifications in the paper restrict attention to observations in which exactly one
of the three “winners” was a firm.
Brehm and Lewis (forthcoming) first shows that the vast majority of leases won by individuals are even-
tually reassigned, relative to roughly half of leases won by firms. These reassignments preview the paper’s
next result: the impact of a firm, rather than an individual, winning the lottery on drilling and production
outcomes is estimated to be a precise zero. These results are consistent with the lease reassignment market
being effective at reallocating leases from low-productivity lessees to high-productivity lessees.
Brehm and Lewis (forthcoming) then studies an interesting subset of the data: leases sold by lottery that
are near a lease that was already producing. In these leases, Brehm and Lewis (forthcoming) finds that when
a firm wins the lottery, the lease is actually less likely to be drilled (by about 10 percentage points) and less
likely to be productive (by about 4 percentage points). That is, leases that are close to nearby oil and gas
production turn out to be more productive when an unqualified individual wins the lottery.
Brehm and Lewis (forthcoming) shows that this empirical result can be explained using a model of
private information. If the newly-sold lease is best developed by the firm operating the nearby parcel,
the fact that that firm likely has private information about the new lease can be a barrier to exchange.
However, if the winner of the lottery is an unqualified individual with a reservation value of zero for the
lease, the tremendous surplus gain that would be realized from trade overcomes the informational barrier,
and the lease is successfully reassigned to the neighboring firm. Consistent with this story, Brehm and Lewis
(forthcoming) finds that reassignment to the nearby firm is significantly more likely if the lottery is won by
an unqualified individual than by another firm.
The upshot of Brehm and Lewis (forthcoming) is then that when a potentially productive lease is held by
an agent that is completely incapable of developing it, the reassignment market is effective at reallocating the
parcel to the most productive user. However, in the presence of private information—which is pervasive in
settings where nearby tracts have already been developed—trade may fail to reallocate the lease to the most
productive user if the current owner is reasonably productive, even if it isn’t the very best firm to develop the
lease. These results highlight that the allocation created by the primary leasing stage—when mineral owners
initially lease their mineral rights to firms—can have long-run impacts on allocative efficiency despite the
existence of a secondary lease market. This implication is consistent with the large efficiency differences
between privately negotiated and publicly auctioned leases documented in Covert and Sweeney (2019).
2.2.7 Opportunities for future work on oil and gas leasing
IO researchers have made remarkable progress in recent years in deepening our understanding of U.S. oil
and gas leasing. Much of this progress has benefited from new datasets covering state and private leasing
activity. Nevertheless, there remain unanswered questions. We see the following topics as especially likely
to be fruitful opportunities for future research:
1. The leasing process is initiated in a variety of ways. Many states and the Bureau of Land Manage-
ment (which manages federal onshore oil and gas leasing) ask firms to nominate tracts for regularly-
scheduled lease sales, and then auction off a subset of those tracts. In the federal OCS and some other
states, area-wide leasing is used in which firms can bid on any of a large number of tracts in a large
27
area. And in the private market, leasing is typically initiated in a decentralized manner when a firm
approaches a mineral owner. A handful of papers have explored differences between area-wide leas-
ing versus nominations (Hendricks and Porter, 2007) and unstructured (private) vs structured (public)
leasing (Covert and Sweeney, 2019), but more work is needed to understand the implications of these
differences in how tracts are selected for leasing.
2. Recent work has found evidence supporting both a common-values framework for lease auctions
(see, e.g. Bhattacharya et al. (2020)) and evidence supporting private values (Covert and Sweeney,
2019). While oil and gas leasing is traditionally viewed as a canonical example of a common values
setting, characteristics of the shale revolution, in which different firms may use different fracking and
completion techniques (see section 2.4 below), suggest that private values are likely to be important.
One valuable agenda for future work would therefore be to develop econometric models of auctions
in which bidders’ values have both common and private value components.
3. The literature on oil and gas lease auctions ubiquitously models the mineral owner as trying to max-
imize some convex combination of revenue, drilling activity, and total (owner + firm) surplus. How-
ever, some mineral-owning governments are increasingly concerned about the greenhouse gas emis-
sions associated with oil and gas produced from public lands and waters. Research is needed to under-
stand how leasing policies—including not just the bonus, royalty, and primary term clauses discussed
above but also novel tools such as fixed per-barrel fees—might impact leasing, drilling, and emissions,
and how such policies might interact with other policy instruments like broad cap-and-trade programs
or carbon taxation.
2.3 Information spillovers and externalities
In many situations, firms control leases that lie within the same or nearby pools of oil and gas. Thus, the
production from wells drilled on one lease is likely to be correlated with the production from wells drilled
on another. This correlation implies that drilling a well generates value not just from the well’s actual
production but also from the information that is revealed about the pool’s quality. The upshot is then that
forward-looking firms should consider these information spillovers when making decisions about when to
drill exploratory wells.
This subsection discusses how IO economists have modeled firms’ strategic exploration incentives given
the likely importance of information spillovers in practice. We begin by discussing models of situations in
which different firms own nearby leases in the same pool. In this situation, information externalities imply
that firms’ drilling timing decisions should not be modeled as a single-agent dynamic problem but rather as
a dynamic game. This body of work generally shows that “free-riding”—delaying investments to wait and
see how well other firms’ wells do—is a common behavior in exploratory drilling.
These findings on free-riding are important not just for their direct application to oil and gas exploration
but also because they connect to broader a literature on free-riding in research and development. See, for
instance, the fairly general model of experimentation and free-riding proposed in Bolton and Harris (1999).
28
The normative implication is that the overall level of knowledge-generating investment by firms is likely
less than that which a full-information, surplus-maximizing social planner would implement.
2.3.1 A theoretical framework for free-riding and the incentive to delay exploration: the “war ofattrition”
Economists’ understanding of information externalities in oil and gas exploration was first developed in
Hendricks et al. (1987), Hendricks and Kovenock (1989), and Hendricks and Porter (1996). Our discussion
of this work will focus on Hendricks and Porter (1996), which posits a model similar to that in Hendricks
and Kovenock (1989) and expands on the empirical work in Hendricks et al. (1987). We will walk through
the model in Hendricks and Porter (1996) in some detail because it is sufficiently simple that it leads to an
analytic solution to firms’ dynamic drilling game that nicely highlights the mis-coordination and free-riding
problems that are present.
Hendricks and Porter (1996) considers a model of drilling in what the paper terms an area-cohort of
leases: leases that are in a geologically similar area and were awarded in the same auction (and therefore
all expire on the same date, five years after the auction). Each tract has a deposit whose size is given by
eX , where X is normally distributed with mean θ and a normalized standard deviation of 1. Capturing the
idea that leases in an area-cohort are geologically similar, θ is common within an area-cohort, but its value
is unknown by the firms.
If a firm drills an exploratory well on its lease, it incurs a cost c and then its deposit size x is publicly
revealed. The value of a deposit is given by π(x), which is equal to zero for deposits too small to be
commercially viable (“dry holes”). The model abstracts away from time-varying oil prices or drilling costs,
so this value is independent of when the firm drills.
Each lease i has T periods in which it may be drilled, where a “period” should be thought of as the time
required to make and then execute an exploratory drilling decision (about three months, according to the
paper). In the first period, the only information available to each firm about its lease is the signal si that is
revealed by the auction (which is not modeled). si is defined to have a normal distribution with a mean of
xi and precision τi. The density function for firms’ beliefs about θ is then normal, with mean µ that is a
weighted average of the si, and a precision ρ, where the weights and ρ are functions of the τi. As firms drill
their leases, revealing the xi, Hendricks and Porter (1996) then applies Bayes rule to show that the posterior
mean of θ will then shift to be a weighted average of the revealed xi with the signals si on the undrilled
leases. Moreover, the beliefs about the size of the deposit on each undrilled lease i will also shift, becoming
a weighted average of the posterior mean of θ with the initial signal si.
Hendricks and Porter (1996)’s model therefore generates an option value of waiting to drill—even
though the oil price is effectively constant—because each firm obtains payoff-relevant information from
observing the drilling outcomes of other firms. So when should each firm drill? If a firm’s signal si is suffi-
ciently high and precise, the firm should drill immediately rather than wait, since other firms’ outcomes are
unlikely to change its beliefs. But what if several firms each control marginal tracts, for which new infor-
mation could plausibly shift beliefs about whether the tract is profitable to explore or not? In this situation,
firms’ drilling problem becomes a dynamic game in which firms would like to wait and gain information
29
from other firms’ outcomes, but face risk in doing so since those firms may choose to wait as well.
Hendricks and Porter (1996) at this point focus attention on a simple case of their model in which there
are just two firms, i and j. The paper first shows that the subgame in which one firm has already drilled is
straightforward: the remaining firm has a belief about its deposit size that is informed by the other firm’s
outcome, and then given that belief it is optimal to either drill immediately or not drill at all (since no
other variables are time-varying). The subgame for period T in which neither firm has yet drilled is also
straightforward, since in that case each firm drills if its belief about its deposit size (based on the initial
signals si and sj) is such that drilling is expected to be weakly profitable.
Specifically, let Vit denote firm i′s expected payoff from drilling its lease in period t. Vit will be a
function of the firm’s belief about xi, and in either of the two subgames discussed above, firm i should drill
if Vit > 0. But what about subgames in which neither firm has yet drilled, and the final period T has not
yet been reached? Here, Hendricks and Porter (1996) proceed by backward induction. Consider period
t = T − 1. Let Wi,T−1 denote the expected payoff, in period T , to i from waiting one period and then
responding optimally to firm j’s drilling outcome. Note that we must have Wi,T−1 > Vi,T−1 since there is
some (perhaps quite small) probability that firm j’s outcome will be so bad that it is optimal for firm i to not
drill at T . In addition, let qj,T−1 denote the probability firm j drills during period T − 1. Firm i’s expected
payoff from waiting is then given by β(qj,T−1Wi,T−1 +(1−qj,T−1) max{0, Vi,T−1}), where β is the firms’
common discount factor, and the expression takes advantage of the fact that Vi,T = Vi,T−1 if firm j does not
drill in period T − 1.
Firm i is then indifferent between drilling in period T−1 (and obtaining max{0, Vi,T−1}) versus waiting
in period T − 1 iff qj,T−1 is given by q∗j , as defined by:
q∗j =(1− β) max{0, Vi,T−1}
β(Wi,T−1 −max{0, Vi,T−1})(11)
Equation (11) tells us that the probability q∗j of firm j drilling that is required to make firm i indifferent
in period T − 1 decreases with the discount factor β, so that firm i is more likely to drill immediately if
it is impatient. Moreover, the higher is firm i’s belief about its deposit size, then the higher is Vi,T−1, the
closer Vi,T−1 is to Wi,T−1, and the greater is q∗j (and q∗j may exceed 1, in which case firm i drills for sure
and obtains expected payoff Vi,T−1).
A remarkable feature of the model in Hendricks and Porter (1996) is that q∗j is constant for all periods
t = 1, 2, ..., T − 1. To see this result, consider period T − 2. Because there are no time-varying variables
in the model, it must be that Wi,T−2 is the same as Wi,T−1. For the same reason, firm i’s value Vi,T−2 from
drilling in period T − 2 must be the same as the value Vi,T−1 from drilling in T − 1 in the event that firm j
has not yet drilled. Thus, the value q∗j required to keep firm i indifferent in period T − 2 must be the same
as in T − 1. This argument can then be repeated all the way back to period t = 1.
The upshot of Hendricks and Porter (1996)’s model is then that, barring a situation in which one of
the firms has a signal that is so strong that it drills in period 1 for sure, the two firms play a mixed strategy
equilibrium that can be characterized as a war of attrition, “since the payoffs from following (letting the other
firm drill first) exceed the payoffs from leading” (p. 393). This equilibrium results in a host of inefficiencies
30
relative to the optimal program that would have been implemented by a single owner of both leases. In that
program, any drilling is always completed by the second period, and the decision to drill at least one well
accounts for the information that will be generated regarding the profitability of drilling a second well. In
the non-cooperative equilibrium, however, drilling may be delayed all the way to the last period, the leases
may be drilled simultaneously rather than sequentially, and they may even be drilled in the wrong order.
Hendricks and Porter (1996) then show that the delays predicted by their model are borne out in data
from “wildcat” (exploratory) leases in the federal OCS that were awarded from 1954–1979. Hendricks and
Porter (1996) defines 270 area-cohorts based on lease sale dates and administrative geographic classifica-
tions. Within each area-cohort, Hendricks and Porter (1996) then computes quarterly exploratory drilling
hazards. Consistent with the model, these hazards exhibit a strong U-shape, with many leases drilled either
in the first few quarters or just before lease expiration. In addition, whether a tract is drilled promptly is
positively correlated with a high winning bid for that tract. There is some suggestive evidence that drilling
(and successful drilling) on other tracts increases the probability of drilling a given tract, but these results are
unfortunately quite noisy. One institutional problem the paper notes is that, while drilling activity is clearly
publicly available information, production outcomes may not be (at least not without a significant lag).
2.3.2 Empirical evidence on cross-firm information spillovers and strategic delay
Since Hendricks and Porter (1996), a suite of papers have examined empirically whether, and how strongly,
the “war of attrition” affects exploratory drilling in practice. Levitt (2016) studies the determinants of wild-
cat drilling in Alberta, Canada, using comprehensive data from 1930–2005. Using fixed effects regressions
to try to control for unobserved firm-by-play heterogeneity, Levitt (2016) finds that a firm’s exploration rate
increases with both the number of successful exploration results it had in the past and the number of success-
ful results obtained by other firms. The latter, spillover effect is roughly an order of magnitude smaller than
the own-firm effect, but it is nonetheless consistent with information spillovers per a model like that of Hen-
dricks and Porter (1996). An important identification assumption underlying a causal interpretation of the
result in Levitt (2016) is that there are no time-varying unobserved factors (e.g., technology improvements)
that affect firms’ exploration success rates.
In contrast to the empirical work in Hendricks and Porter (1996) and Levitt (2016), Lin (2013) and
Hodgson (2018) develop and estimate structural models of firms’ dynamic exploration game. The mod-
els used in these papers are richer than that of Hendricks and Porter (1996) in a variety of ways, thereby
increasing the models’ fidelity to the setting at hand but imposing the cost that the model must be solved
computationally rather than analytically.
Lin (2013), like Hendricks and Porter (1996), uses data from the federal OCS. Lin (2013)’s model
specifies that the profits a firm earns from drilling are directly affected by neighboring firms’ exploratory
and development drilling. The main objective of Lin (2013) is to estimate the signs and magnitudes of the
parameters governing these effects, using data on drilling activity and production outcomes.
Hodgson (2018), in contrast, studies oil and gas exploration in the UK portion of the North Sea during
1964–1990, with the objective of estimating counterfactuals that quantify how the “war of attrition” delays
drilling and reduces industry surplus. The model in Hodgson (2018) is explicitly spatial. The paper begins by
31
estimating the spatial correlation of exploration success across blocks using a Gaussian process regression,
finding that the probability of successful exploration in any given block is informed by outcomes occurring
within one or two neighboring blocks, but not further than that. Motivated by this correlation, the paper then
runs regressions similar to those in Levitt (2016), finding that a firm is more likely to explore a particular
block if nearby blocks experienced successful exploratory drilling, regardless of whether the success was
accomplished by the same firm or a different firm (and controlling for firm, block, and time fixed effects).
To quantify the extent to which free-riding dampens drilling, Hodgson (2018) then develops and esti-
mates an econometric model of firms’ exploration decisions. Rather than make each firm’s profits a direct
function of other firms’ drilling, as in Lin (2013), Hodgson (2018) builds a model in which the relevant state
variable for each firm in each period (month) is its belief—as generated by the Gaussian process model given
the industry’s exploration history—about the probability of exploration success on each block. The model
simplifies firms’ action space each period by specifying that each firm can explore in and then develop at
most one block in each period. The structure of the game, like the simpler analytic game in Hendricks and
Porter (1996), gives firms an incentive to delay exploration and development in the hope of free-riding on
information revealed by other firms’ drilling.
In both Lin (2013) and Hodgson (2018), firms make drilling decisions as part of a dynamic exploration
game, the solution concept for which is Markov perfect equilibrium. Specifically, each period each firm
makes exploration and development decisions based on publicly available information and its own private
information about tract-level profitability. Both papers estimate their models’ structural parameters using a
two step procedure; Lin (2013) uses the estimator from Pakes et al. (2007), while Hodgson (2018) adopts
the Bajari et al. (2007) estimator. The core assumption used in both papers to identify information spillover
effects is that there are no time-varying unobserved factors that are geographically correlated and affect the
probability of successful drilling.
Lin (2013) finds no evidence supporting significant externalities from exploratory drilling. This result
echoes the noisy correlations found in Hendricks and Porter (1996) between drilling and outcomes on nearby
tracts in the federal OCS. Lin (2013) does, however, find that profits from drilling are positively, and signif-
icantly, increased after a neighboring firm drills a development well. A natural interpretation of this result
is that drilling a development well is a strong signal that the deposit contains a sufficiently large quantity of
reserves that development is profitable.
Hodgson (2018) uses its estimated model to simulate a counterfactual that shuts down the incentive
to free ride by forcing firms to believe (incorrectly) that other firms will never drill. Doing so increases
the number of blocks developed by 1990 by 28%, and industry profit increases by 31%. These substantial
effects highlight the importance of strategic interactions in governing firms’ exploration behavior, and they
illustrate how these interactions can substantially reduce exploration investment and firms’ surplus.
Taken as a whole, this literature has developed a useful theoretical framework for understanding the
incentives generated by information externalities in oil and gas exploration, and it has taken initial steps at
empirically quantifying the importance of these effects in practice. The strongest evidence for significant
drilling delays comes from the North Sea (Hodgson, 2018), whereas the evidence from the Gulf of Mexico
is more mixed (Hendricks and Porter, 1996; Lin, 2013).
32
Much remains to be learned on this topic, particularly from the onshore U.S. where the shale boom
is taking place. In addition, the work discussed in this sub-section has generally taken lease assignments
as given. However, a firm’s decision to place a high bid on a particular tract may be related to its beliefs
about when (or whether) other firms will develop their leases. Beyond inducing a form of selection into
firms’ participation in a dynamic drilling game, firms’ anticipation of how the game will play out may
enter into their behavior during the auction, and thereby impact the sellers’ revenues. The implications
of links between lease auctions and strategic interactions during subsequent drilling could be developed in
future research. Such work would of course be useful for understanding leasing and drilling behavior in this
important industry, and at the same time it may also help shed light on investment games in other settings
where knowledge spillovers are important.
2.3.3 Exploration spillovers across tracts governed by different extraction policies
The papers discussed thus far in this subsection have studied situations in which different firms control
leased exploration tracts, but the mineral owner of all the tracts is the same (a sovereign government). In
the onshore United States, however, it is commonly the case that mineral ownership of a pool is fragmented
between the federal government, state governments, and private entities. These different owners will typ-
ically write leases with different terms and have different policies that govern how wells may be drilled.
This “patchwork” of fragmented ownership and governance policies within a pool, within which production
outcomes are likely to be geologically correlated, leads to a situation where one owner’s policy choices can
affect outcomes for other owners. This form of policy spillover is related to but distinct from the economics
of the dynamic drilling games discussed above. It is also a form a spillover that has parallels in settings
beyond oil and gas exploration. In the pharmaceutical industry, for example, different countries can im-
pose different policies (patents, subsidies, etc.) to support research and development. Because capital can
flow between countries, and because innovations can often be applied globally, changes to one country’s
innovation policy can affect research expenditures elsewhere.
Lewis (2019) examines the effects of patchwork state and federal mineral ownership using data on ex-
ploration, drilling, and production in Wyoming. Wyoming is an enticing setting for this question due to
the peculiar way in which land is often divided between federal and state owners.13 Typically, comparing
drilling and production outcomes on federal versus state land would be problematic because differences in
ownership may have been selected based on underlying geological productivity. Lewis (2019) takes advan-
tage of the fact that, in many places in Wyoming, the pattern of state versus federal ownership is effectively
random due to the Land Ordinance Act of 1785, which established a regular land grid and allocated 2 of
every 36 square mile sections to the state.
Lewis (2019) begins by developing a model of firms’ exploratory and development drilling decisions
when facing land in which drilling costs are heterogeneous, motivated by the idea that federal land is likely
to be more costly than state land due to federal environmental regulations. A key feature of this model
is the fact that the underlying oil and gas productivity on any given section of land is correlated with the
productivity of nearby sections. Thus, when drilling costs are relatively low on state land, exploratory13The remoteness and harsh climate of the region studied in Lewis (2019) has led to very little private ownership.
33
drilling on state land increases not just because of the direct cost reduction, but also because exploration is
diverted away from nearby federal land. Lewis (2019) then shows that this prediction is borne out in the
data: exploratory drilling is particularly likely on state land but particularly unlikely on federal land near to
state land.
Lewis (2019) also considers how spatial cost heterogeneity affects the drilling of development wells
(which follow exploratory wells if they are successful) and ultimate oil and gas production. Here, while the
theoretical predictions are ambiguous, Lewis (2019) again finds evidence of increased drilling and produc-
tion on state land, and reduced drilling and production on federal land close to state land (relative to federal
land far from state land). It is further noteworthy that Lewis (2019) shows that the differences in drilling and
production activity on state versus federal land do not materialize strongly until the 1970s, coinciding with
the establishment of the EPA and passage of the Endangered Species Act.
A key implication of Lewis (2019)’s findings is that tighter environmental restrictions on federal land
cause a non-trivial amount of leakage of drilling and production activity onto nearby state land. Thus, any
regulatory cost-benefit analysis that looks only at federal land will fail to account for both the substitutability
of nearby state land (which reduces the regulations’ costs to firms) and the additional environmental damages
occurring on state land (which reduces the regulations’ environmental benefits).
2.4 Productivity, innovation, and learning
The most important driver of the shale oil and gas revolution has been the remarkable increase in the produc-
tivity of shale wells. In this subsection, we discuss research that has endeavored to understand the sources of
productivity growth in the oil and gas industry. Some of these studies have defined and studied “productiv-
ity” in the way economists have traditionally done when studying other industries: a Hicks-neutral additive
production shifter in an equation specifying the determinants of output. Other work, however, has shed
light on forms of productivity improvements—for instance, how firms learn about the production function
itself—that have garnered less attention in the economic literature. Together, this body of work has sub-
stantive importance for understanding the shale boom itself, and for understanding its future trajectory as
oil and gas compete with renewable sources of energy that are themselves becoming less costly to produce.
In addition, this work has value for understanding productivity dynamics more broadly, particularly as the
well-level investment and output data that are accessible to researchers in the oil and gas business typically
do not have an equivalent in most other industries.
2.4.1 Learning about the production function and optimal input choices
Economic models of firms’ production and input mix decisions often assume that firms are flawless profit-
maximizers and cost-minimizers, given the input and output prices that they face. But for new technologies,
the form of the production function may initially be unknown to firms, and it must be instead learned through
experience. In such situations, firms’ initial input choices will not be consistent with profit-maximization, or
even with cost-minimization conditional on output. Moreover, the process of learning about the production
function may then be an important mechanism for increases in firms’ output, efficiency, and profits.
34
Covert (2015) makes some of the very first progress on evaluating this learning mechanism by studying
firms’ fracking input choices, and the resulting production outcomes, in the Bakken shale oil formation in
North Dakota. As with essentially all modern shale wells, wells drilled in the Bakken are “fracked” by
injecting large volumes of water, sand, and other fluids into a well at high pressure. Doing so fractures the
oil-bearing shale in the vicinity of the wellbore, releasing the oil so that it can be produced. The amount of
water and sand used, along with other details of the fracking process, are important determinants of each
well’s ultimate production.
To understand the novel form of learning studied in Covert (2015), consider a simplified version of the
paper’s production function, in which Qi denotes well i’s output, Wi is injected water, Si is injected sand,
and lati and loni denote the well’s latitude and longitude:
logQi = f(lati, loni) + βWi logWi + βSi logSi + εi (12)
A standard econometric evaluation of productivity growth would estimate equation (12) and then mea-
sure increases in the total factor productivity residuals εi over time. The premise of Covert (2015) is instead
that an important way firms increase output Qi (and profits) over time is by learning about the true values of
βWi and βSi , and then making better input choices.
The first step in Covert (2015) is to estimate the “true” production function using data on all wells drilled
during the paper’s 2005–2012 sample. The production function model, and the Gaussian process regression
used to estimate it, allows the marginal impacts of water and sand, given by βWi and βWi in equation (12)
above, to vary flexibly with each well’s geographic location. This flexibility turns out to be important, since
the estimates imply that the shape of the production function—and the output-maximizing volumes of water
and sand—vary substantially over space. The identification assumption is that, conditional on flexible spatial
controls, water and sand inputs are exogenously determined (by, for instance, firms’ experimentation with
different input mixes).
Once the true production function is estimated, Covert (2015) moves on to study how the profitability
of firms’ input choices, given the estimated production function, has varied over time. Doing so requires
knowledge of the costs of additional water and sand inputs. Covert (2015) obtains this information using
data on drilling and completion costs that are available for a subset of wells in the dataset. Covert (2015)
finds that in 2005, at the start of the shale boom, firms’ input choices are far lower than those that would
maximize profits. The average share of profits firms obtain versus what they could have obtained if they
chose the optimal input levels is just 21%. The share of profits obtained steadily rises over time as firms
use more water and sand in their frac jobs, reaching 60% by 2012. Thus, Bakken firms were able to achieve
large increases in profits captured (and production) over seven years by increasing their use of inputs.
Covert (2015) also considers the extent to which firms make good input choice decisions, given the
information they have available at the time they drill. That is, for a well drilled on a particular date t, Covert
(2015) creates an estimated production function that just uses information from wells prior to t. In this
framework, firms’ input choices look initially less bad in the early part of the sample. However, differences
between the optimal and actual sand and water use increase over time, as optimal sand and water use increase
more quickly than actual sand and water use.
35
Covert (2015)’s results suggest that firms are not making optimal use of the data on outcomes from
previously drilled wells that they have available to them. The paper finds that one possible mechanism for
this result is that firms focus too heavily on information from their own wells rather than wells drilled by
other firms. The paper arrives at this conclusion by showing that firms’ input choices are consistent with a
model in which each firm estimates a production function in which its own wells receive a weight that is
larger than wells drilled by others. Covert (2015) also finds that firms have an aversion to experimenting,
since they tend to avoid input choices for which the current-information production function provides a noisy
estimate of what the outcome may be. This lack of experimentation may be responsible, at least in part, for
why the industry overall was performing below the optimal frontier even in the last year of the sample.
Nonetheless, Covert (2015) shows that even with this shortfall, improved input choices drove substantial
production and profitability growth in the Bakken shale over 2005–2012.
Another implication of the results in Covert (2015) is that firms considering completing wells in shale
formations potentially benefit from waiting for other firms to move first, so that they can learn more in
advance about the production function. That is, there is a free-riding incentive based not just on learning
about the size of the underground deposit—as in the papers discussed in subsection 2.3—but also on learning
about the production function.
These free-riding incentives are considered in Steck (2018), which also uses the Bakken shale as its
empirical setting. At its core, Steck (2018) combines the production function estimation of Covert (2015)
with the estimation and simulation of a dynamic investment game, as in Hodgson (2018) and Lin (2013).
Similar to Covert (2015), Steck (2018) finds that wells’ input use and production have increased considerably
over time, concluding that firms were learning about the production function. This evidence of learning then
motivates a dynamic model in which firms choose both when to drill a well on each of their leases and the
inputs with which to frack it. The publicly-observable state in each period includes parameters that dictate
how production is likely to respond to increased input use. Firms rationally update their beliefs about these
parameters in response to observed input use and production outcomes from wells drilled by other firms.
Steck (2018) finds that, overall, the incentive to learn from other firms’ input choices is not strong enough
to substantially delay drilling. This finding may be related to the fact that Covert (2015) does not find that
firms experiment a great deal, so there is limited value to a firm from waiting to see what other firms do.
Steck (2018) does find that, in situations where firms are highly uncertain about the production function—for
instance, when few wells have been drilled overall—the delay incentive can become economically large.
In a similar vein, Fetter et al. (2020) examines the extent to which firms learn from observing other
firms’ choices of fracking inputs, focusing on chemical use in the Marcellus Shale in Pennsylvania. The
paper examines a policy change that required firms to publicly disclose the chemicals they include in their
fracking fluid, starting on April 2012. Fetter et al. (2020) shows that after the policy change, firms’ fracking
“recipes” became more similar to one another, using a Jaccard similarity index to measure the similarity
between the high-dimensional vectors of chemical concentrations between any given pair of wells. Fetter et
al. (2020) also finds that experimentation with new chemical mixes falls following the 2012 policy change.
36
2.4.2 Learning where to drill
The papers discussed in section 2.4.1 above point to better input selection as an important driver of pro-
ductivity growth in shale oil and gas. Agerton (2020) examines another potentially important factor: firms
learning about where to drill, not just how to drill. This idea, in other settings, would be analogous to firms
becoming better at selecting good investment opportunities, not just better at executing them.
The paper studies the Haynesville Shale in Louisiana, in which individual leases are force-pooled into
administrative units that are typically one square mile (and in practice line up with Public Land Survey
System sections). These units are sufficiently large that firms can drill, frack, and produce from more than
one well. The analysis in Agerton (2020) centers on firms’ decision of whether, after drilling the first well,
they should drill additional wells.
In Agerton (2020)’s model, firms have an initial, noisy signal of how productive the pooling unit is likely
to be. Firms will generally find it advantageous to drill at least one well, even if this signal is somewhat
poor, since doing so reveals information and indefinitely holds the leased acreage in the entire unit. Given
the information revealed from this first well, firms can then decide whether to drill subsequent wells.
Agerton (2020) solves its model by backwards induction. The drilling decision for subsequent wells is
an infinite horizon problem, and the drilling decision for the initial well is a finite horizon problem in which
drilling results in an immediate (and possibly negative) expected payoff from drilling and in unlocking the
continuation value associated with future drilling. The key parameters of the model—drilling costs, time-
varying technological productivity and drilling costs, and the noisiness of the initial signal—are estimated
via the nested fixed point algorithm of Rust (1987), using data on both drilling timing and wells’ output.
Agerton (2020) does not explicitly model fracking input choices, so improvements in input use are incorpo-
rated into the model’s time-varying technological productivity term. To avoid the complexity of modeling
a dynamic game, Agerton (2020) assumes that information obtained from drilling does not spill over across
firms.
Agerton (2020) finds that changes in firms’ drilling location selection over time explain most of the
productivity improvements in the Haynesville shale over 2008–2016. In the early years of the Haynesville’s
development, firms often drilled wells in poor locations, both to hold acreage and because they were not
fully informed about local reservoir quality. Subsequently, as most pooling units became held by their initial
well, drilling activity shifted to follow-up drilling, but only in units that were highly productive. The upshot
is that a naive estimator that doesn’t account for this effect would find that technological progress improved
wells’ productivity by 7% per year. After accounting for site selection, however, technological progress
accounts for productivity improvements of just 2% per year.
2.4.3 Productivity in vertical relationships
A feature of many industries—and especially oil and gas—is that production requires contributions from
multiple firms that are vertically linked to one another. In oil and gas, the exploration and production
companies that acquire leased acreage and organize the drilling and production process actually outsource
the drilling of wells to specialized firms that own and operate drilling rigs. Because data are available on both
37
the production companies and drilling companies (and indeed the individual rigs) responsible for drilling a
given well, the oil and gas industry offers an opportunity to empirically understand productivity dynamics
in vertical relationships.
First, Kellogg (2011) studies data from Texas during 1991–2005, mostly pre-dating the shale boom. The
paper’s measure of productivity for a given well is the number of days required to drill it. The idea is that an
important driver of drilling cost is the time it takes to drill, and in the absence of comprehensive drilling cost
data, drilling time is a good proxy for drilling costs. Kellogg (2011) then models drilling time as a function
of the producer’s own experience, the rig’s own experience, and the joint experience between the producer
and the rig. To address the possibility that producers and rigs that are especially good matches may work
together more frequently, biasing the estimate of the effect of producer-rig experience, the model includes
fixed effects for each producer-rig pair (along with field fixed effects to control for geologic heterogeneity
and time fixed effects to control for technological progress).
Kellogg (2011) finds that relationship-specific learning is an economically important driver of produc-
tivity improvements. Rigs that stick with one producer can be expected to improve their productivity (i.e.
reduce their drilling times) twice as quickly as rigs that frequently change producers. The average real-
ized benefit from relationship-specific learning in the data is a nearly 5% reduction in drilling times, which
Kellogg (2011) estimates to be worth $12,400 in cost savings per well.
In addition to relationship-specific learning, vertical relationships can also drive variation in produc-
tivity over time if the quality of matches between firms varies over time. Vreugdenhil (2020) studies this
possibility in the context of producer-rig matching in the U.S. Gulf of Mexico during 2000–2009. In this
setting, wells vary in the level of their complexity. For instance, some wells are deeper than others or may
require horizontal drilling techniques. Rigs also vary in their ability to drill and complete relatively deep
or complex wells. Vreugdenhil (2020) groups rigs into low, medium, and high efficiency based on their
depth ratings: the maximum water depth at which they can operate. The paper then shows that, on average,
high-efficiency rigs tend to be matched to relatively complex wells. This positive assortative matching is
intuitively efficiency-enhancing since the capabilities of high-efficiency rigs are not needed for simple wells.
Vreugdenhil (2020) then documents an intriguing pattern: the strength of positive assortative matching
co-varies positively with oil and gas prices. That is, wells and rigs are likely to have better matches during
a price boom than a price bust.
To explain this boom-bust pattern, Vreugdenhil (2020) proposes a model of costly search in which new
wells (“projects”) enter the market each period and must search to find a rig. The paper models search as
flexibly falling on a spectrum between fully random and perfectly targeted. The probability of successfully
contracting a rig is modeled as an increasing function of the ratio of available rigs to the number of searching
projects. This ratio falls during booms and rises during busts.
Upon being contacted by a project, the rig can decide whether to accept the match—in which case the
price is determined by Nash bargaining—or wait until the next period. This accept / reject decision then
gives rise to the covariance between match quality and boom / bust periods. In a boom, rigs have a strong
incentive to pass on poor matches because they have a high probability of being matched in the following
period. Thus, matches that are formed are likely to be of high quality. In busts, however, rigs are willing to
38
accept poor matches because the probability of being contracted by a project in future periods is low.
2.4.4 Lessons learned and paths for future work on oil and gas productivity
Understanding productivity growth in the oil and gas business is important for its own sake. The U.S.
shale oil and gas boom was a direct consequence of productivity improvements, and the extent to which
productivity continues to improve will help determine the extent to which zero-emission energy sources can
successfully compete with oil and gas as substitute fuels. In addition, the wealth of data on investments
into individual wells, and on wells’ ultimate output, has allowed researchers to empirically examine novel
mechanisms for productivity growth, such as improvements in input selection. These mechanisms are likely
important in other industries but difficult to empirically evaluate because typical datasets aggregate data to
the establishment-year level rather than providing data at the level of individual investments.
The most impactful way that IO researchers can build off the progress made thus far would be to investi-
gate whether the ideas and tools used to study oil and gas productivity growth can be applied to technologies
like wind, solar, and geothermal energy production. Improvements in these industries’ productivity is crucial
to future decarbonization. And like oil and gas, these industries feature data on the location and timing of
individual investments, so there is potential for the ideas and tools that have been used to study productivity
in the oil and gas sector to be transferable to these newer, zero-emission technologies.
2.5 Environmental regulation of resource extraction and transportation
Every stage of the fossil fuel industry—including extraction, transportation, and final consumption—is as-
sociated with emissions of local pollutants and greenhouse gases. Tightening regulation of these emissions
is arguably the largest issue facing this industry going forward. We highlight in this section how indus-
trial organization research has helped shed light on the impacts (both prospectively and retrospectively) of
environmental regulations on resource extraction and transportation.
2.5.1 Regulation of environmental damage at production sites and of site decommissioning
One set of important environmental regulations concerns preventing local harm in the vicinity of production
sites. We already discussed in section 2.3.3 one paper—Lewis (2019)—that shows how regulations gov-
erning drilling on federal lands decreased federal exploratory drilling but increased exploratory drilling on
adjacent state land. These effects amount to a form of leakage of drilling activity and (potentially) pollution
away from a stringently regulated area to a less-stringently regulated area.14
Another recent paper considering leakage is Vreugdenhil (2021), which studies how deepwater drilling
rigs respond to increased stringency of regulations aimed at reducing the risk of oil spills and blowouts in the
Gulf of Mexico (like the BP Deepwater Horizon disaster in 2010). In this setting, the potential for leakage
is international, as deepwater rigs can travel across the ocean to other basins. The main goal of Vreugdenhil14In related work in the downstream refining sector, Sweeney (2015) discusses evidence of leakage and environmental policy
spillovers in the context of reformulated gasoline standards that only apply in some U.S. gasoline markets.
39
(2021) is to evaluate the potential leakage from U.S. spill prevention policies that (per engineering estimates)
would increase the cost of drilling by 20% while decreasing spills per well drilled by 20%.
To assess this counterfactual, Vreugdenhil (2021) adopts most of the model from Vreugdenhil (2020)
(discussed in section 2.4 above). To that earlier framework, Vreugdenhil (2021) adds a model of rigs’
location choice. In this discrete choice model, the value of moving to a location j is given by the expected
price-cost margin at j, minus the cost of moving to j, plus an idiosyncratic type I extreme value error term.
The value of the moving cost is proportional to distance, with the constant of proportionality given by data
on rates charged by marine transport ships. Vreugdenhil (2021) then estimates location-specific drilling
costs and the scale of the error term by fitting the discrete choice model to empirical choice probabilities.
The assumption needed for identification of the scale term—which dictates how responsive rig moves are
to price or cost changes in the long-run—is that there are no unobserved time-varying cost changes across
location.
After estimating the model, Vreugdenhil (2021) finds in its main policy counterfactual that for every
1 barrel reduction in U.S. oil spills, movement of rigs away from the U.S. causes 0.48 barrels of spills
elsewhere in the world. Most of this leakage is driven by the most technically sophisticated rigs, which are
estimated to be especially sensitive to the regulation and are associated with larger wells and higher spill
volumes. Leakage is not 100% for two reasons. First, rigs are not perfectly price sensitive, leading some rigs
to stay in the U.S. market even though U.S. margins are depressed, relative to elsewhere in the world, in the
long run. Second, the crowding of rigs into other regions decreases their utilization rates. Still, Vreugdenhil
(2021)’s estimate of 48% leakage is large, especially considering that the paper does not incorporate any
mechanism for leakage via increases in oil prices.
While Lewis (2019) and Vreugdenhil (2021) are primarily concerned with environmental hazards during
the drilling or operation of oil and gas wells, wells at the end of their economic life—when their production
rate has declined so far that production revenue no longer covers fixed operating and maintenance costs—
also present an environmental challenge. Such wells must be properly decommissioned, not just simply shut
in by closing the surface valves, in order to avoid risks from leaks of methane (a potent greenhouse gas) or
toxic chemicals into the local environment (Environmental Protection Agency, 2021; Raimi et al., 2021).
This decommissioning process typically involves installing a plug in the wellbore and filling it with cement,
which can cost $20,000 alone before any required surface reclamation (Raimi et al., 2021).
These decommissioning costs are problematic because, in the absence of strong policy incentives, firms
will avoid or postpone decommissioning. Muehlenbachs (2015) illustrates this problem using data from
84,000 wells in Alberta, Canada and a dynamic model of firms’ decision whether to continue operating a
well, temporarily shut it in, or permanently decommission it. In this model, temporarily shutting in a well
preserves the option value of resuming production in the future should oil prices increase. The firm values
this option, but crucially does not value the externality imposed by the presence of a non-decommissioned
wellbore. Meanwhile, the Alberta government requires decommissioning but does not impose a deadline on
firms, who can leave wells temporarily shut-in indefinitely. These circumstances lead Muehlenbachs (2015)
to examine firms’ incentives to leave wells in a temporarily shut-in state rather than decommissioning them.
A simplified version of the model in Muehlenbachs (2015) is as follows. If a well is producing in a
40
given period, it earns profits of (P − C)Q, where P is the oil price, C is the marginal extraction cost,
and Q is the production quantity (which declines over time). A well that is temporarily shut-in carries an
inactivity cost ofM , and a decommissioned well carries no annual cost. The model includes switching costs
for transitions between any of these states, leading to hysteresis in firms’ decisions. Decommissioning is an
absorbing state, and the switching cost into this state represents the decommissioning cost.
Muehlenbachs (2015) estimates its model using the Rust (1987) nested fixed point method. The amount
of hysteresis in the data pins down the switching costs, and then the extent to which firms actually decom-
mission wells is informative about the costM of maintaining a well in a temporarily shut in state. The paper
then uses the estimated model to shed light on why so many wells are temporarily shut in rather than decom-
missioned. The key counterfactual finds that even if natural gas prices were doubled, the number of active
wells would increase by just 6% (via switching out of the temporarily shut-in state). This result suggests
that the main value firms derive from temporarily shutting in wells is not the option value of restarting them
in the future, but rather the value of avoiding decommissioning costs.
A seemingly natural policy solution to the problem of avoided decommissioning would be to force
firms to promptly decommission wells following a period of inactivity, or to expose them to full liability
for the environmental harms from leaking, non-decommissioned wells. These solutions are unfortunately
confounded by difficulties. The liability-centered solution is only effective if gas and fluid leaks are de-
tectable and attributable to specific wellbores, which is challenging in practice. And both solutions face
the “judgment-proof problem” (Shavell, 1986) that decommissioning and environmental liabilities can be
avoided via bankruptcy. This problem may be especially severe in the oil and gas industry, where wells that
are nearing the end of their productive lives can be sold to poorly-capitalized firms who are unlikely to be
able to fund decommissioning.
The classic solution to the judgment-proof problem is to require firms, prior to drilling, to post a bond
that is large enough to cover the well’s decommissioning. Yet in the U.S., state and federal bonding re-
quirements have generally been far short of expected decommissioning costs, leading to a proliferation of
“orphaned” wells that do not have a solvent owner (Ho et al., 2018).
Would increasing bonding requirements help slow the creation of orphaned wells? The evidence pre-
sented in Boomhower (2019) suggests that the answer is yes. Boomhower (2019) studies a substantial
increase in Texas’s bonding requirements that was enacted in 2001. This increase led to: (1) the sale of
a large number of wells from small operators with poor environmental records to larger firms; (2) a 70%
decrease in the number of new orphaned wells; and a 25% reduction in violations of state water protection
rules.
The economically large effects documented in Boomhower (2019) provide the first large-scale evidence
of the effectiveness of bonding requirements that we are award of. Beyond highlighting the value of in-
creased oil and gas well bonding requirements for jurisdictions outside of Texas, these results also speak
to the likely importance of decommissioning bonds for investments in other assets that may create environ-
mental liabilities absent decommissioning and site remediation.
41
2.5.2 Regulation of emissions from hydrocarbon transportation
The majority of the oil and gas produced in the shale boom is located in rural areas, so that long-distance
overland transportation is required to bring these products to market. Considerable popular and policy
attention has been drawn to increases in the volume of crude oil transported by rail, which carries the risk
of explosive derailments in populated areas (such as the disastrous explosion in Lac-Megantic in 2013) and
is associated with considerable local air pollution from locomotive emissions (Clay et al., 2019).
At the same time, several long-distance oil pipeline projects have drawn attention and concerns regarding
both the potential for local spills and the pipelines’ effects on upstream production and total CO2 emissions.
The Keystone XL pipeline (from Alberta’s oil sands to the U.S. Midwest) and the Dakota Access Pipeline
(DAPL, from the Bakken shale in North Dakota to the Gulf of Mexico) have been especially controversial
and extensively litigated.
These policy debates provoke the question of how regulation of pipeline construction or crude-by-rail
transportation might affect production and transportation of crude oil, along with the associated emissions.
Covert and Kellogg (2018) considers this issue in the context of the Bakken shale, DAPL, and crude-by-
rail. The paper develops a model that captures the essential characteristics of pipeline vs rail transportation,
with the goal of quantifying the degree of substitution between these two modes. Pipeline transport has
nearly zero marginal cost but large up-front sunk construction costs, and construction financing typically
requires that prospective pipeline customers (“shippers”) commit to long-term capacity payments, whether
they actually use the capacity or not. These commitments can be as long as 10 years; thus, to spur pipeline
construction shippers must commit to the line even though oil prices over such a long time horizon are
highly uncertain. Rail transport, in contrast, requires much smaller up-front commitments but does involve
non-trivial marginal costs. In addition, rail shipments can be directed to multiple downstream destinations,
whereas the pipeline is locked into its terminal once built. The trade-off between pipeline and rail trans-
portation is then that rail transport offers shippers greater temporal and spatial flexibility, but at a higher
per-barrel cost.
The simple version of the model in Covert and Kellogg (2018) is as follows. Let the upstream supply
curve be given by the function Q(P ), and suppose pipeline capacity is equal to K. If the downstream
price Pd is relatively low, then that will induce little upstream supply, so that Q < K, and with zero
marginal transportation cost on the pipeline, the upstream and downstream prices will be equal. But for
high downstream prices, enough supply is induced that the pipeline flows at capacity (Q = K), and there
can be a wedge between the upstream and downstream prices. It is this price wedge that is the reward to
pipeline shippers who committed to pipeline capacity in advance. The magnitude of the price wedge is
limited, however, by the ability of shippers to move crude-by-rail at a marginal cost r. Were the price wedge
to exceed r, crude-by-rail flows would increase, pulling up the upstream price and (after a modest lag due
to crude-by-rail adjustment costs) eventually equating the price wedge to r.
The willingness of pipeline shippers to commit to constructing a pipeline of capacity K is then de-
termined by the expected price wedge that will be realized during the duration of the commitment. This
expected value is an increasing function of the cost r of crude-by-rail and a decreasing function of K. Thus,
regulations that increase the cost of crude-by-rail lead to increases in capacity investment.
42
Covert and Kellogg (2018) estimates each of the components of its model using monthly data on oil
prices, crude production and flows, and costs of rail transportation. The paper’s main results then stem
from the finding that pipeline and rail transport of crude oil are highly substitutable. An increase in rail
costs of $2 per barrel is estimated to motivate an increase in DAPL’s capacity of 12–29%, depending on
the specification. This substitution implies that the elasticity of rail flows with respect to rail costs is quite
large: -0.9 to -2.2. Conversely, if DAPL’s construction had been blocked, then crude-by-rail would replace
82–91% of the lost pipeline flows.
The upshot of Covert and Kellogg (2018) is then that environmental regulations targeting just one trans-
portation mode will primarily displace crude oil transportation to the other mode, rather than reduce the
overall volume of crude oil transported. Such substitution can be beneficial if the regulated mode is as-
sociated with substantially greater environmental externalities than the other mode. However, if the main
objective of the regulation is to reduce overall crude oil production, then regulating just one of these two
closely substitutable transportation modes is unlikely to be successful.
2.5.3 Interactions between downstream environmental regulations and market power in resourcetransportation
Reducing emissions from coal-fired electric power generation is essential for controlling both local pollution
and CO2 emissions. A series of papers has demonstrated, however, that the exercise of market power
by railroads—which are responsible for nearly all U.S. long-distance coal movements—can substantially
complicate the impacts of downstream environmental regulations.
Busse and Keohane (2008), Preonas (2019), and Hughes and Lange (2020) all document evidence that
railroads engage in substantial price discrimination on coal movements.15 In Busse and Keohane (2008),
the basis for price discrimination is the enhanced regulation of sulfur dioxide (SO2) emissions per the Clean
Air Act Amendments (CAAA) of 1990. The CAAA specified that a subset of coal power plants (“table A”
plants) were newly required to participate in an SO2 emission allowance trading market, giving them strong
incentives to reduce their emissions. One way to do so was to invest in scrubbers—an expensive undertaking.
The other option was to switch coal supply to low-sulfur coal in the Powder River Basin (PRB) of Wyoming.
This latter option required railroad transportation, and essentially all table A plants were connected to the
PRB by just one or two potential rail carriers.
Using data on generators’ coal procurement costs, Busse and Keohane (2008) shows that railroads en-
gaged in price discrimination in response to the imposition of the SO2 trading program. In particular, deliv-
ery costs for PRB coal increased for table A plants relative to non-table A plants, and these cost increases
were especially large for table A plants that were relatively close to the PRB.
The implication of Busse and Keohane (2008)’s findings is that, in response to a regulation that substan-
tially increased the value of PRB coal, railroads were able to take advantage of their market power to capture
that value, at the expense of generators (and likely, ultimately ratepayers). Preonas (2019) and Hughes and
Lange (2020) in essence consider the reverse of this situation: the decline in generators’ value of coal caused
by the shale boom’s dramatic expansion of natural gas supply. The reduction in the price of natural gas over15Hughes (2011) similarly finds evidence consistent with price discrimination on ethanol movements.
43
the past decade has enabled gas-fired electric generators to sell power at low prices, curtailing the revenues
that coal-fired generators can earn. Both Preonas (2019) and Hughes and Lange (2020) exploit hetero-
geneity in coal plants’ exposure to natural gas competition to show that more exposed plants experienced
significant decreases in their coal input prices, consistent with railroad price discrimination. Preonas (2019)
shows that coal price decreases were especially large for “captive” plants reliant on only a single railroad,
while Hughes and Lange (2020) finds especially large coal price decreases for plants exposed to deregulated
wholesale power markets, where competition with gas-fired generators would be most severe.
Preonas (2019) goes on to draw an analogy between its results and the potential impacts of carbon
pricing in the electricity sector, which (like a decrease in the price of natural gas) would disadvantage coal-
fired generators. Preonas (2019) points out that if an analyst were to forecast changes in emissions from such
a tax under an assumption that coal were priced at marginal cost, the analyst would likely over-estimate the
extent of coal-to-gas switching—and therefore emissions reductions—that would be caused by the policy.
The upshot of this set of papers is then that the impacts of environmental policies that impact fuel use at
the point of consumption (here, electric power generators) can be influenced substantially by firms (here,
railroads) who are upstream in the supply chain.
2.5.4 Opportunities for future work on impacts of environmental regulation on fossil fuel extractionand transportation
Industrial organization economists have made excellent progress in understanding the behavioral responses
of resource extraction and transportation firms to environmental regulations. We view this literature as still
being in an early stage of development, and the likelihood of continued increases in regulatory stringency
(particularly regarding greenhouse gas (GHG) emissions) makes this area ripe for future work. In particular,
we see the following areas as especially high-potential for research contributions:
• Methane emissions. Methane is an especially potent greenhouse gas, and oil and gas extraction has
been identified as an important source of methane emissions. Methane emission control is bedeviled
by the difficulty of leak detection and substantial heterogeneity across firms and sites in leak rates.
IO research can be valuable in designing and evaluating regulatory schemes, and the enforcement of
such schemes, that aim to reduce methane leaks.
• Leasing policy and climate change. Policies that put a price on carbon emissions will interact with
oil and gas leasing, which historically has put an implicit price on oil and gas production, via the
imposition of royalties. In addition, the response of lease terms to carbon prices will determine how
the loss of producer surplus is split between mineral owners and firms, and the extent to which carbon
pricing reduces drilling and production activity versus reducing the royalties and bonuses imposed by
mineral owners.
• Applications to zero-emissions energy sources. Are economic models and lessons learned from the
oil and gas sector applicable to zero-emissions technologies that, like oil and gas, rely on geographic-
specific inputs? Like oil and gas, large-scale geothermal, wind, and solar energy production all require
44
contracts with landowners, and as was the case with the shale boom, productivity improvements
will be central to the success of these technologies. As data on these energy sources becomes more
widespread, application of models and research designs originally conceived for oil and gas are likely
to be fruitful in understanding the economics and these zero-emissions technologies.
3 Personal transportation, energy use, and environmental regulation
This section discusses the industrial organization of transportation, energy use, and the environment. We
focus our attention on how ideas and methods from industrial organization have helped (and going forward,
can continue to help) answer questions about the impacts of policies aimed at reducing emissions from
the transportation sector. These policies include fuel taxes, fuel economy standards, regulation of emis-
sions of local pollutants, and incentives for adoption of alternative fuel vehicles—all of which have been
implemented or received serious consideration in a variety of jurisdictions.
While our discussion will emphasize the substantive contributions of industrial organization to these
policy-relevant topics, we also note that work in this area has also made positive contributions to broadly
relevant IO topics and methods, including estimation of consumers’ valuation of goods’ attributes, models
for how firms set product characteristics in differentiated oligopoly markets, models of consumers’ search
behavior in retail markets, and evaluation of the impacts of incentives and standards in network goods
markets (here, electric vehicles (EVs) and EV charging stations). The contributions we discuss also relate
to the IO literature on automobile markets that endeavors to understand consumers’ vehicle demand, firms’
behavior, and market equilibrium. We will not extensively discuss this broader literature here—much of it
is covered in other chapters of this handbook volume—though many of the papers we discuss will use tools
from that literature, especially the differentiated products demand and Bertrand oligopoly model in Berry
et al. (1995). Because the methodological details behind implementing these tools are discussed elsewhere
in this volume, here we will emphasize how these tools enable papers’ substantive contributions rather than
devote significant time to the tools themselves. That said, we will highlight instances where papers at the
intersection of IO, transportation, and the environment augment IO methods with new features that are
important for capturing phenomena relevant to the question at hand. For instance, we will discuss how the
constraints imposed by fuel economy standards affect firms’ pricing decisions in Bertrand competition, as
modeled in Jacobsen (2013).
The section proceeds in five subsections: (1) consumers’ demand for fuel economy and associated
implications for fuel economy policy; (2) economic impacts of fuel economy standards; (3) regulating
emissions of local air pollutants; (4) consumer search in retail gasoline markets; and (5) markets for EVs
and EV charging stations.
3.1 Estimating consumers’ demand for fuel economy, and implications for fuel economypolicy
Vehicle fuel economy is an important vehicle attribute that consumers value. The distance a particular
vehicle can be driven per dollar spent on gasoline (miles per dollar) was included as a demand-side attribute
45
in the original Berry et al. (1995). Vehicle fuel economy has also long been a focus of policy attention, given
interest in reducing the externalities associated with gasoline consumption.
Policy-makers have a variety of tools at their disposal to improve vehicles’ fuel economy. Two prominent
such tools have been gasoline taxes and fuel economy standards. Gasoline taxes have long been upheld by
environmental and public finance economists as the most efficient approach for reducing greenhouse gas
emissions from automotive fuel consumption. As Anderson and Sallee (2016) notes in a recent review of
this literature, an important reason behind economists’ favorable view of gasoline taxes is that that they
provide incentives to reduce fuel use via not just improving vehicles’ fuel economy, but also by reducing
miles traveled.16
Anderson and Sallee (2016) also notes, however, that gas taxes may not be fully effective if consumers
fail to fully internalize future fuel costs when they purchase vehicles. That is, if consumers are myopic
and under-value future fuel costs relative to the up-front vehicle price, then their response to a gasoline tax
(in terms of purchasing more efficient vehicles) will be attenuated relative to what a model assuming fully
rational, forward-looking behavior would predict. An argument for fuel economy standards is then that they
are not adversely affected by this consumer “myopia” because they can directly mandate a high level of
fuel economy. Moreover, under this argument fuel economy standards can be viewed as actually improving
consumers’ welfare—in a paternalistic sense—since they can correct the “internality” caused by consumers’
myopic decision-making.
An essential input to the discussion of fuel taxes versus fuel economy standards is then empirical evi-
dence on the extent to which consumers actually value future fuel costs when they purchase a vehicle. This
subsection discusses a suite of papers that employ demand models to answer this question, using a variety
of research designs. Before discussing individual papers, however, we think it will be useful to highlight
several of the empirical challenges that must be overcome (or assumptions that must be made) in order to
credibly address the question.
To begin, consider the following model of consumers’ utility from purchasing a vehicle, letting i denote
a consumer and j denote a particular model:
Uij = −αpj − γFj + βXj + ξj + εij (13)
In equation (13), pj denotes the vehicle’s price, Fj denotes its expected lifetime fuel costs, and Xj
denotes other vehicle characteristics. ξj then denotes an unobserved vertical product characteristic, and εijis an idiosyncratic error term. The goal is to test whether γ is equal to α. If that null hypothesis is rejected,
and instead γ < α, we would take that as evidence that consumers are myopic. And in that case, the ratio
γ/α tells us how many cents on the dollar consumers value future fuel costs relative to the up-front vehicle
purchase price.
One immediate obstacle to estimating equation (13) is that doing so requires a computation of expected16Anderson and Sallee (2016) offers a brief survey of the literature that estimates the elasticity of gasoline consumption or vehicle
miles traveled with respect to the price of gasoline. This literature can rightly be described as enormous, and it spans the fieldsof environmental economics, public finance, macroeconomics, and industrial organization. We therefore do not cover gasolinedemand estimation in this chapter but rather direct readers to Anderson and Sallee (2016) and the reviews and papers cited therein.
46
future fuel costs (at the time of vehicle purchase) Fj . Computing Fj requires estimates of how many miles
the vehicle will be driven each year, the vehicle’s expected remaining life before scrappage, the vehicle’s
fuel economy, expected future fuel prices, and the consumer’s discount rate. None of these components
is straightforward to obtain or compute, and the papers we describe below use a mix of approaches and
datasets, as we will discuss.
Second, the test of consumer myopia as we described it above assumes that consumer tastes—or at
least the parameters α and γ—are homogeneous. All of the papers in this literature make this assumption
(sometimes implicitly, sometimes explicitly).17 Doing so certainly adds value in terms of yielding a sharp
answer to the research question, but of course in reality tastes or the degree of myopia across consumers
may vary. In that case, the object of interest would be the distribution of the relevant preference parameters,
and therefore myopia, across the population.
Finally, identification of γ in equation (13) requires an unconfoundedness assumption that future fuel
costs Fj are not correlated with unobserved vehicle characteristics that constitute ξj . An important compo-
nent of Fj is the vehicle’s fuel economy, which will typically be correlated with a variety of other perfor-
mance characteristics of the vehicle. Because available and operationalizable data on vehicle characteristics
are typically limited—often to variables like horsepower, weight, acceleration, and footprint—this uncon-
foundedness assumption can be difficult to defend. Thus, papers in this literature have tended to adopt
empirical strategies that isolate the variation used to identify γ to sources other than cross-sectional varia-
tion in vehicles’ fuel economy.
3.1.1 Identifying consumers’ valuation of fuel costs from used vehicle prices
This subsection discusses three papers closely-related papers—Busse et al. (2013), Allcott and Wozny
(2014), and Sallee et al. (2016)—that study consumers’ valuation of fuel economy by using data on used
vehicle markets, and leveraging used vehicle price variation induced by shocks to the price of gasoline. All
three of these papers conclude that consumers fully, or nearly fully, value future fuel use when they are
purchasing a vehicle.
A motivation for studying the used car market is that the supply of a particular vehicle model by model-
year (for instance, a 1996 Honda Accord) can be approximated as fixed. With fixed supply and with homo-
geneous preferences for vehicle attributes, the preference parameters in equation (13) can then be estimated
directly using the hedonic pricing equation (14):
Pj = −γαFj +
β
αXj + ξj (14)
The benefit of this hedonic approach is that it is not necessary to estimate consumers’ price sensitivity α,
so that the standard equilibrium price endogeneity problem of demand estimation can be avoided.18 There
are at least two caveats to the “fixed supply” assumption underlying this approach, however. First, to the17Grigolon et al. (2018), which we discuss below, does somewhat weaken this assumption by modeling heterogeneity in con-
sumers’ vehicle miles traveled.18Busse et al. (2013) includes analyses of new car transactions, in which the paper estimates fuel use valuation after imposing a
new car demand elasticity from the literature. The paper then concludes that buyers of new cars also fully value future fuel use.
47
extent that consumers substitute between used cars and new cars, quantity changes in the new car market will
affect equilibrium outcomes in the used car market. Second, any price-sensitivity of the vehicle scrappage
rate (which Jacobsen and van Benthem (2015) finds evidence for) will violate the assumption. Allcott and
Wozny (2014) relies on the hedonic model for its main estimates but explores the potential importance of
both of these caveats in an appendix, finding that its results do not change substantially.
To address potential omitted variables biased caused by unobserved vehicle characteristics that are cor-
related with fuel economy, each of Busse et al. (2013), Allcott and Wozny (2014), and Sallee et al. (2016)
adopts a strategy of including detailed vehicle model by model-year (or in the case of Allcott and Wozny
(2014) model by model-age) fixed effects in its pricing equation and then leveraging variation in future fuel
costs that is induced by changes in fuel prices. Thus, these papers estimate equations like (15) below, where
t indexes the month-of-sample, and µj denotes the model by model-year (or model-age) fixed effects:
Pjt = −γαFjt + µj + τt + εjt (15)
Equation (15) includes month-of-sample fixed effects τt to capture market-wide changes in vehicle
prices (and preference shifts to or from the outside good), and the error term εjt accounts for period-to-
period changes in model-specific tastes. Model j’s lifetime fuel costs Fjt are now a function of fuel prices
at t, and the underlying fuel price variation is assumed to be uncorrelated with εjt.
To illustrate the calculation of Fjt, we follow Allcott and Wozny (2014); other papers in this literature
make similar calculations. Allcott and Wozny (2014) obtains information on vehicle miles traveled (VMT),
differentiated by vehicle class and age, from the National Household Travel Survey (NHTS). It obtains
survival probabilities from vehicle registration data. The paper uses the Environmental Protection Agency
(EPA) combined (city and highway) fuel economy rating for each vehicle. For a discount rate, Allcott and
Wozny (2014) averages interest rates associated with equity returns and vehicle loans, in order to capture
the opportunity cost of funds for cash transactions and financed transactions, respectively.
The final component of Fjt is the expected path of gasoline prices over the lifetime of the vehicle, which
is not necessarily the same as the gasoline price at the time of the vehicle’s purchase. One approach to model-
ing future expected fuel prices—and that used by Allcott and Wozny (2014) in its headline specification—is
to use futures market prices. Another is to use survey data on consumers’ beliefs from the Michigan Survey
of Consumers, which Anderson et al. (2013) finds correspond closely to a belief that gasoline prices are a
martingale.
Allcott and Wozny (2014) finds that consumers modestly under-value future fuel costs, estimating a
value of γ/α = 0.76 in its baseline specification. In alternative specifications this value ranges from 0.46 to
1.01, though the full valuation result requires a large (15%) assumed discount rate. Busse et al. (2013) finds
results that generally imply full valuation. The difference in findings between Allcott and Wozny (2014) and
Busse et al. (2013) is not large and may be attributable to either their use of different datasets (Allcott and
Wozny (2014) uses prices from wholesale transactions while Busse et al. (2013) uses retail transactions) or
differences in their specifications.19
19In addition to the difference in model by model-year versus model by model-age fixed effects noted above, Busse et al. (2013)differs from Allcott and Wozny (2014) in that it replaces Fjt with an interaction between vehicle fuel economy and the gasoline
48
Finally, Sallee et al. (2016) also finds results consistent with consumers fully valuing future fuel costs,
but identifies γ/α using a different source of variation: heterogeneity in odometer readings within vehicles
of a particular model by model-year. For instance, consider two 2002 Toyota Priuses (a very efficient model),
one of which has a lot of miles on it and the other very few. Following a gasoline price increase, we would
expect the price of the vehicle with fewer miles on it to be more responsive to the gas price shock, since it
is likely to have a longer remaining life. Sallee et al. (2016) isolates this variation in the data by carefully
computing Fvjt in a way that accounts for the odometer reading of each particular vehicle v, and by including
fixed effects for all interactions between model, model-year, and month-of-sample. The identifying variation
in Sallee et al. (2016) is therefore more restricted than that used in Allcott and Wozny (2014) and Busse et
al. (2013), but nonetheless Sallee et al. (2016) finds a similar headline result: approximately full valuation
of future fuel costs.
3.1.2 Consumer valuation of fuel costs for new vehicles
An important limitation of the studies discussed in section 3.1.1 above is that they only speak to consumers’
behavior when purchasing used vehicles, not new vehicles. Because new vehicles are, by definition, only
new for one year, research designs based on leveraging only variation in fuel prices over time (e.g., by
including model by model-year fixed effects in the estimating equation) will not work when only data from
new vehicle markets are used. So alternative research designs are necessary to study consumer behavior in
these markets.
Grigolon et al. (2018) studies new vehicle transactions in seven European countries during 1998–2011.
The paper leverages the fact that even within a given model by model-year, there exists variation in vehicles’
fuel economy because consumers can choose between different engine types. In particular, many models
are offered with both diesel and gasoline engine types, and diesels achieve considerably better fuel economy
than do gasoline-powered vehicles (at the cost of a higher sticker price). At the same time, Grigolon et al.
(2018) also allows for heterogeneity in consumers’ VMT, so that consumers who anticipate driving more
will be more likely to select relatively efficient vehicles.
Grigolon et al. (2018)’s embeds this within-model variation in fuel economy into the following model
of consumers’ indirect utility function:
uijkt = −αpjkt/yt − γρβmi ejktgkt/yt + xjktβxi + ξj + ξt + ξjkt + εijkt, (16)
where i indexes consumers, j indexes vehicle models, k indexes engine variants, and t indexes country by
month-of-sample. pjkt denotes the vehicle’s sticker price. The marginal utility of money is assumed to be
inversely proportional to income, so the utility loss from paying for vehicles and fuel is scaled by market-
level income yt. The second term on the right-hand-side of equation (16) denotes discounted expected future
fuel costs, calculated as the product of the consumer’s expected annual VMT βmi with the vehicle’s fuel use
per mile ejkt, fuel prices (either gasoline or diesel) gkt, and a capitalization coefficient ρ ≡∑S
s=1(1 + r)−s
price in the main estimating equation. Then, after estimation, Busse et al. (2013) transforms the parameter estimate to an implieddiscount rate using information on VMT and scrappage rates.
49
(where S is the vehicle’s expected lifetime).
As with the literature discussed in section 3.1.1 above, a goal of Grigolon et al. (2018) is to test whether
α = γ. The crucial identification assumption is that each vehicle’s fuel economy is uncorrelated with
the vehicle’s other unobserved characteristics. To help buttress this assumption, the utility specification
(16) includes other observed characteristics xjkt (such as horsepower and size), model fixed effects ξj , and
market fixed effects ξt. The identification assumption is that other vehicle-specific unobservables ξjkt are
uncorrelated with inverse fuel economy ejkt and other characteristics xjkt.
Grigolon et al. (2018) estimates equation (16) using the methods and instrumental variables discussed
in Berry et al. (1995), with the twist that the distribution of consumers’ VMT βmi is restricted to follow the
distribution of reported VMT from survey data. It finds an estimate of γ/α = 0.91, consistent with nearly
full valuation of future fuel costs by consumers. This result is robust to alternative specifications that shut
down preference heterogeneity over vehicle characteristics and fuel use.
Grigolon et al. (2018) also leverages the estimated heterogeneity in VMT to make an important point
about the efficacy of fuel taxes versus product taxes (or fuel economy standards). Fuel taxes induce high-
VMT consumers to differentially select efficient vehicles, whereas product taxes do not. The model in
Grigolon et al. (2018) allows the paper to quantify this effect, finding that a fuel tax reduces total fuel use
by 18%, relative to a reduction of just 12% from a revenue-equivalent product tax.20
Another recent paper that studies the new vehicle market, but comes to a different conclusion, is Gilling-
ham et al. (forthcoming). This paper studies the U.S. market, exploiting variation in fuel economy ratings
induced by a 2012 scandal in which two automakers, Hyundai and Kia, were caught overstating the fuel
economy of several of their top-selling vehicle models. As one of several consequences, the U.S. EPA
required Hyundai and Kia to restate the fuel economy of these vehicles. Gillingham et al. (forthcoming)
exploits the changes this restatement had on new vehicle prices and sales volumes—using transaction-level
sales data—to estimate consumers’ valuation of future fuel costs.
Gillingham et al. (forthcoming) begins by providing narrative evidence that the fuel economy restate-
ment was sudden and likely unexpected by consumers. It then estimates the effect of the restatement on
equilibrium vehicle prices using a panel fixed effects (difference-in-difference) regression that focuses the
identifying variation on changes in prices for affected versus unaffected models produced by Hyundai and
Kia. The result is that the restatement reduced the price of affected vehicles by 1.2%, equal to about $300 on
average. The paper invests considerable effort in showing that this estimate is primarily driven by decreases
in the price of affected vehicles rather than increases in the price of these vehicles’ close substitutes.
Based on estimates of VMT, future fuel prices, expected vehicle lifetimes, and discount rates that follow
the assumptions used by the papers discussed in section 3.1.1, Gillingham et al. (forthcoming) finds that
this sticker price decrease is substantially less than change in the affected vehicles’ future fuel expenditures.
With a 4% discount rate, consumers are estimated to value future fuel costs by only 17% of how they value
the up-front purchase price.
An alternative explanation for this under-valuation result is that the quantity supplied of the affected
vehicles also declined, attenuating the decrease in the equilibrium price. However, Gillingham et al. (forth-20To compute these counterfactuals, Grigolon et al. (2018) assumes Bertrand competition by automakers, per Berry et al. (1995).
50
coming) estimates a small and statistically insignificant positive effect of the restatement on equilibrium
quantities: +5% with a standard error of 4%. Moreover, even if one postulates that quantities actually de-
creased by 5% (5 times the size of the estimated price decrease), consumers estimated valuation of future
fuel costs doubles at most.21
3.1.3 Lessons learned and paths forward for research on consumers’ valuation of fuel economy
The extent to which vehicle consumers value future fuel expenditures when making purchase decisions is a
crucial input for evaluating fuel economy policy. Using a variety of demand models, estimation methods,
and datasets, research on this question has made substantial progress over the past 15 years. Three papers—
Busse et al. (2013), Allcott and Wozny (2014), and Sallee et al. (2016)—all find that in the U.S. used vehicle
market, consumers fully or nearly fully value future fuel costs at the time of purchase. Grigolon et al. (2018)
finds the same for the European new vehicle market, but Gillingham et al. (forthcoming) arrives at a starkly
different result—substantial undervaluation—for the U.S. new vehicle market.
A potential explanation for the discrepancy between Gillingham et al. (forthcoming) and the other papers
we discussed is that the population of new car buyers in the U.S. behaves differently than the population of
U.S. used car buyers (or European new car buyers). Another is that consumers are more attuned to changes
in gasoline prices or differences in engine types than they are to changes in fuel economy ratings. Yet another
is that the undervaluation result is simply specific to the case examined in Gillingham et al. (forthcoming):
a scandal-induced fuel economy restatement affecting two automakers. Future research is needed to help
distinguish between these explanations for these papers’ diverging results.
In addition, more work is needed on heterogeneity in consumers’ valuation of fuel economy. With the
exception of Grigolon et al. (2018)—which allows for valuation heterogeneity that is induced by hetero-
geneity in VMT—all of the papers discussed in this section use models in which consumers’ valuation of
fuel economy is homogeneous. As pointed out in Anderson et al. (2013), consumers’ valuations are likely
to be heterogeneous not just due to VMT, but also due to heterogeneity in local gasoline prices, beliefs about
future gasoline prices, discount rates, and inattention. The welfare effects of fuel economy policies depend
on this heterogeneity and which mechanisms lie behind it, so research is needed to better understand it.
Finally, it would be useful to extend the literature on consumers’ valuation of future fuel costs to study
alternative fuels, and in particular electric vehicles (EVs). One component of the appeal of EVs is that the
price per unit energy for electricity is likely to be lower than that for gasoline. Even if the literature that
studies consumers’ valuation of fuel economy for gasoline-fueled vehicles reaches a consensus, it is not
obvious that consumers’ behavior when choosing between vehicles with high versus low future gasoline
expenditures will translate directly to choices between gasoline-powered versus electric-powered vehicles.
Research is needed to help inform how variation in the relative prices of electricity and gasoline will affect
consumer adoption of EVs.21To complete the calculation, Gillingham et al. (forthcoming) also uses estimates of consumers’ demand elasticity from the
literature.
51
3.2 Economic impacts of, and firms’ responses to, fuel economy standards
Many countries have tightened their fuel economy regulations over the past decade, often with reducing
CO2 emissions as the stated main policy objective. Understanding the cost-effectiveness and distributional
impacts of fuel economy standards, especially relative to fuel taxation, is essential for informing fuel econ-
omy policy-making. One input for understanding these impacts is consumers’ valuation of fuel economy,
as discussed in subsection 3.1 above. That input alone is not sufficient, however, as the effectiveness of fuel
economy standards will also be heavily influenced by firms’ behavioral responses.
Ideas and methods from industrial organization—especially differentiated products demand and Bertrand
competition models related to Berry et al. (1995)—have proven themselves useful for research into how
firms react to fuel economy standards. This subsection discusses the progress made by this literature. We
begin by discussing Jacobsen (2013), which illustrates how binding fuel economy standards alter vehicle
manufacturers’ pricing decisions relative to what would be predicted by a standard Bertrand pricing model.
We then discuss a series of papers that follow up on Jacobsen (2013) by exploring other mechanisms with
which firms can comply with fuel economy standards, with particular emphasis on the trade-offs firms face
between improving fuel economy versus improving other vehicle characteristics.
3.2.1 Fuel economy standards and automakers’ pricing and fleet mix decisions
Fuel economy standards reduce fuel use and CO2 emissions by setting a ceiling on the fuel use per mile of
newly sold vehicles. When the U.S. first introduced its Corporate Average Fuel Economy (CAFE) standard
in 1978, compliance was evaluated at the original equipment manufacturer (OEM) level. That is, the sales-
weighted average fuel use per mile within an OEM needed to fall weakly below the constraint. Since 2009,
regulations have permitted CAFE compliance credit trading across firms.
Jacobsen (2013) develops a model for how the pre-2009 CAFE standards affected OEMs’ pricing de-
cisions, equilibrium outcomes in vehicle markets, and the surplus received by consumers and producers.
The paper’s model is rich, as it includes models of consumers’ demand and driving behavior, Bertrand price
competition among OEMs, and the used car market. Our discussion will refrain from extensively discussing
all of these components but rather focus on the paper’s key innovation regarding modeling and estimating
the impacts of binding CAFE constraints on the market’s Bertrand pricing equilibrium.
Perhaps most importantly, Jacobsen (2013) accounts for the fact that the U.S. CAFE standard, in the
absence of credit trading across firms, had different impacts on three distinct types of firms. First, some
OEMs, such as Honda and Toyota, produced vehicles that were so efficient that the standard was not binding.
Second, another group of firms, such as BMW and Mercedes, chose not to comply with the standard and
instead paid fines that were a linear function of their non-compliance. The behavior of these first two groups
of firms had been considered by earlier work (Goldberg, 1998). But Jacobsen (2013) argues that a third
group of firms—consisting of the “Big 3” U.S. automakers Chrysler, Ford, and GM—treated the CAFE
standard as a binding constraint, since violating the standard may have caused them to incur reputational
and political costs that substantially exceeded the explicit fines. Consistent with this view, these three
automakers’ weighted-average fuel economy closely matched the standard throughout the period studied in
52
the paper.22
Jacobsen (2013) shows that these three types of firms face profit maximization problems that differ in
important ways. First, consider firms for whom CAFE does not bind. Jacobsen (2013) models them as
solving a standard differentiated product Bertrand profit maximization problem, per equation (17):
maxpj ,j∈J
∑j∈J
(pj − cj)qj(P ), (17)
where j denotes vehicle models and J denotes the set of vehicles produced by the firm. pj and cj denote
vehicle prices and marginal costs, and qj(P ) denotes the quantity sold of each vehicle as a function of the
full vector P of all vehicle prices in the market (including vehicles sold by other firms).
Now consider a firm that is out of compliance and pays fines. The firm’s problem now includes a fine
term that is proportional to the product of total sales with the difference between the fuel economy standard
d (in miles per gallon) and the firm’s sales-weighted average fuel economyCAFE. Thus, per equation (18),
the firm is effectively taxed on its sales of inefficient vehicles.23
maxpj ,j∈J
∑j∈J
(pj − cj)qj(P )− γ(d− CAFE)∑j∈J
qj(P ). (18)
Finally, consider the “Big 3” firms that treated the standard as a binding constraint. These firms faced
the constrained maximization problem in equation (19) below, where mpgj denotes the fuel economy of a
particular model j:
maxpj ,j∈J
∑j∈J
(pj − cj)qj(P ) s.t.
∑j∈J qj(P )∑j∈J
qj(P )mpgj
− d ≥ 0. (19)
The constraint in equation (19) forces automakers to decrease their sales of their inefficient vehicles and
increase sales of efficient vehicles. To do so, they must set markups on efficient vehicles that are smaller
than what would be optimal in standard Bertrand competition (equation (17)), and markups on inefficient
vehicles that are larger.
Jacobsen (2013) then integrates the firms’ problems (17), (18), and (19) into an equilibrium model of the
vehicle market. The paper adopts a demand model from Bento et al. (2009) that is similar to that in Berry
et al. (1995). It estimates the demand system in a first step and then estimates the supply model using the
FOCs implied by equations (17), (18), and (19). In addition to the usual estimation of each vehicle model’s
marginal cost, this step also involves estimating the shadow value of the CAFE constraint for the three firms
for which it binds. Doing so requires additional restrictions, since for any shadow value there exists a set of
marginal cost values that could rationalize the data. Jacobsen (2013) therefore restricts OEMs’ markups to
be proportional to markups imposed by dealers (which are observable), per an argument from Bresnahan and
Reiss (1985). The extent to which these markups are relatively low for efficient vehicles (versus what would22In practice, the regulation allows some banking and borrowing of credits across years, so these firms do not have to satisfy the
standard every year.23We simplify the exposition here by ignoring the different standards imposed for cars versus light trucks. Sales-weighted average
fuel economy is computed as a sales-weighted harmonic mean of each model’s fuel economy in miles per gallon. During the periodstudied in Jacobsen (2013), the fine parameter γ was equal to 50.
53
be implied by the inverse demand elasticity) is informative about the importance of the CAFE constraint.
Jacobsen (2013) estimates that the shadow values of the CAFE constraint for Chrysler, Ford, and GM
are quite large. For instance, the marginal effect of loosening the standard by one MPG for GM’s passenger
car fleet is estimated to be $438 per vehicle. Note, however, that this shadow value describes the effect of
tightening the standard for just GM, while leaving the standard for other OEMs unchanged. Quantifying the
effects of an overall tightening of the CAFE standard requires a counterfactual simulation.
To assess counterfactuals, Jacobsen (2013) integrates its model of the new vehicle market with a model
of the used vehicle market from Bento et al. (2009), in which vehicles are probabilistically scrapped as
a function of their equilibrium price. Jacobsen (2013) then simulates a one MPG increase in the CAFE
standard. An important feature of these simulations is that the policy change takes over a decade to achieve
its full effect, since its effects percolate slowly through the used vehicle market. Used car fuel economy
improves by less than new car fuel economy, since inefficient used cars have high prices (because they are
scarce) and are scrapped more slowly. The reductions in surplus are initially split nearly evenly between
consumers and producers, but after a decade become overwhelmingly incurred by consumers as new vehicles
percolate through the used market. The producer surplus impacts are borne entirely by the “Big 3” firms
on which the standard binds. Other firms actually benefit because they face weakened competition in the
sub-market for large, inefficient vehicles. Overall, the cost of tightening the CAFE standard is $616 per ton
of CO2 avoided. Taken together, the results in Jacobsen (2013) highlight the value of integrating standard
models of equilibrium in vehicle markets with institutional details—such as different firm types and used
car markets—to capture important implications of fuel economy policies.
3.2.2 Fuel economy standards and vehicle attributes
Most of the modeling in Jacobsen (2013) permits OEMs to respond to CAFE only by changing vehicle
prices and quantities sold. The end of the paper considers an extension in which OEMs can incur additional
costs per vehicle to improve fuel economy, using technology cost curves from the engineering literature.
Allowing for technology adoption in this way reduces Jacobsen (2013)’s estimated cost of tightening CAFE,
from $616 to $222 per ton of CO2 avoided.
This last result from Jacobsen (2013) raises the question of how the full extent of OEMs’ possible
responses to CAFE, including sales-mix shifting, technology adoption, and trading off fuel economy against
other vehicle attributes, affects the policy’s outcomes. The potential importance of attribute trade-offs is also
highlighted by Knittel (2011), which estimates OEMs’ technological frontier that determines the trade-offs
between vehicle weight, engine power, and fuel economy. Knittel (2011) finds that this frontier has steadily
pushed outward over time. It also finds, however, that during 1980–2004 automakers slid their choices along
this frontier so that there were large increases in the size and power of new vehicles sold—for instance, new
vehicle horsepower nearly doubled—but only a 6.5% increase in new vehicle fuel economy.
To further understand the incentives behind attribute trade-offs, and how fuel economy choices affect
vehicle attributes and welfare outcomes, Klier and Linn (2012) and Whitefoot et al. (2017) augment the
Bertrand pricing model from Jacobsen (2013) by endogenizing OEMs’ attribute choices. Doing so intro-
duces additional modeling and estimation challenges. First, instead of representing each vehicle model’s
54
marginal cost by some constant cj , these papers need to estimate marginal costs as a function of attributes
xj . This function can in principle be estimated using firms’ FOCs, via a strategy like that in Fan (2013)’s
study of newspaper markets. However, both Klier and Linn (2012) and Whitefoot et al. (2017) instead use
engineering models to estimate the cost function.
A second challenge is that the standard demand estimation strategy of using non-price characteristics
as instruments is invalid because these characteristics are chosen simultaneously with the unobserved char-
acteristics in ξj . Klier and Linn (2012), for instance, states that “the firm may choose a higher price and
greater horsepower for a vehicle that consumers perceive as being ‘sporty’ or of higher ‘quality”’. Whitefoot
et al. (2017) addresses this problem by partitioning the set of characteristics into those that can be adjusted
in the short-to-medium run and those that are determined by long-run planning schedules, such as vehicle
dimensions, powertrain type (e.g. hybrid vs conventional), and drive type. It then uses the latter set of
characteristics (of both same-manufacturer and different-manufacturer vehicles) as instruments. Klier and
Linn (2012) instead exploits the fact that OEMs often offer models in different vehicle classes (e.g. SUVs
versus full-size sedans) that share an engine platform. The instruments in Klier and Linn (2012) are then
characteristics of same-manufacturer models from different classes but sharing an engine platform. The idea
is that these characteristics will be highly correlated for cost and technology reasons, but the characteris-
tics of vehicles in a different class should not be correlated with the demand shifter ξj of the vehicle under
consideration.
Whitefoot et al. (2017) uses its model to simulate the effects of replacing the 2006 U.S. fuel economy
standards (27.5 mpg for cars and 21.6 mpg for trucks) with the 2014 standard (34.0 mpg for cars and 26.3
mpg for trucks). It finds that OEMs will engage in both sales-mixing and attribute-shifting. For instance,
to meet the tightened requirement the average 0 to 60 miles per hour acceleration time will increase by 0.7
seconds. Unlike Jacobsen (2013), Whitefoot et al. (2017) finds that almost all of the additional compliance
costs will be borne by consumers, whether or not firms are permitted to adjust vehicle attributes (it is not
clear to us why these two papers arrive at different results on CAFE’s cost incidence). But the ability to
adjust attributes decreases the simulated loss of consumer surplus in 2014 from $12.8 billion to $7.4 billion.
Klier and Linn (2012) examines the effects of a one mile per gallon increase in CAFE standard strin-
gency, finding results that are rather different from those in Whitefoot et al. (2017). Klier and Linn (2012)
finds that OEMs constrained by the standard do not respond by decreasing attributes like horsepower or
size, but instead adopt technology that improves fuel economy while holding these other attributes constant.
In equilibrium, the sales-weighted vehicle horsepower and size still decrease, but only because the sales of
large, powerful cars decrease. Furthermore, like Jacobsen (2013), Klier and Linn (2012) finds a more even
split in compliance costs between consumers and producers. Producers are found to benefit substantially
from the ability to adjust vehicle attributes, while consumers actually wind up slightly worse off.
It is not clear why Whitefoot et al. (2017) and Klier and Linn (2012) arrive at such different conclusions.
One possibility is simply that the effects of a marginal change in CAFE standard stringency (as in Klier
and Linn (2012)) are different than those of a large change (Whitefoot et al., 2017). The two papers also
model different vehicle attributes, use different engineering models as a basis for their cost functions, and
use different identification strategies for their demand models. Additional research that explored these
55
differences would be valuable. Moreover, the industry has evolved substantially since these papers were
published, so a refresh with up-to-date data would be useful for informing current fuel economy policy.
Finally, Anderson and Sallee (2011) takes a different approach to estimating the cost to firms of CAFE
standards. Instead of estimating an equilibrium model of the vehicle sector, as in Jacobsen (2013), Klier and
Linn (2012), and Whitefoot et al. (2017), Anderson and Sallee (2011) exploits a “loophole” in fuel economy
regulations together with a condition implied by firms’ profit maximization. The loophole is that, prior to the
Obama revision of fuel economy standards, automakers were able to overstate the fuel economy of vehicles
with technology that enabled them to be “flexible-fuel”, typically meaning that they could operate on fuel
that was 85% ethanol. The cost of this technology is known from engineering estimates to be roughly $100
to $200 per vehicle.
Under a set of assumptions—including that automakers must be at an “interior solution” (i.e., not fully
exhausting the loophole) and that consumers do not directly value flex-fuel vehicles—Anderson and Sallee
(2011) shows that automakers should equate, on the margin, the cost of flex-fuel technology with the cost
of increasing fuel economy. Anderson and Sallee (2011) then argues that these assumptions hold in prac-
tice and concludes that the cost to automakers of complying with CAFE is merely $9–$27 per vehicle.
This estimate is far lower than that from Jacobsen (2013), potentially because automakers can comply by
changing vehicle attributes or investing in technology—channels that are absent from Jacobsen (2013)’s
main estimates. Jacobsen (2013) also notes that Anderson and Sallee (2011) draws on data from a period
when gasoline prices were high (so that consumers were willing to pay for fuel economy anyway), whereas
gasoline prices during most of Jacobsen (2013)’s sample were low.
3.2.3 Attribute-based fuel economy standards
Fuel-economy standards in many countries are “attribute-based”, in that the fuel economy target for a par-
ticular vehicle model is a function of that model’s weight or footprint. Japan and much of Europe use
weight-based fuel economy standards, and since 2008 the U.S. CAFE standard has been footprint-based. Ito
and Sallee (2018) shows that standards structured in this way give automakers an incentive to up-size their
vehicles in order to slacken the fuel economy constraint that they face, leading to a deadweight loss distor-
tion. Ito and Sallee (2018) then examines vehicle characteristics data from Japan, where the fuel economy
target for vehicle models discontinuously decreases at discrete vehicle weights. Ito and Sallee (2018) shows
that the precise weight of Japanese vehicles is tightly bunched just above these discrete points, providing
direct evidence that automakers respond to the incentives created by attribute-based regulation.
3.2.4 Gaming of fuel economy standards
Another way that firms can respond to fuel economy policy is to cheat. Reynaert and Sallee (forthcoming)
documents that, following the enactment of stringent fuel economy regulations in Europe in 2007, a large
gap developed between automakers’ laboratory fuel economy ratings of their vehicles and actual on-road
performance, using driver-level panel data from the Netherlands. In 2014, Reynaert and Sallee (forthcom-
ing) documents an average fuel consumption performance gap that exceeds 50%. To provide a welfare
interpretation of this gaming of the standard, the paper then develops a model that captures automakers’
56
incentives to game and allows for two possible beliefs on behalf of consumers: that they know the gaming is
occurring, or that they are fooled at the time of purchase. Taking the model to the data, Reynaert and Sallee
(forthcoming) shows that even when consumers are fooled (and therefore buy cars that do not maximize their
ex-post utility), consumers still benefit from gaming because the cost reductions from non-compliance with
the true fuel economy standard partially pass through into lower vehicle prices. However, these consumer
surplus gains come at the expense of increased CO2 emissions.
Reynaert (forthcoming) develops a model that incorporates the full spectrum of ways discussed above
that automakers may respond to fuel economy standards: changing their sales mix, adjusting vehicle at-
tributes (downsizing), adopting fuel-saving technology, and gaming. The paper focuses its attention on the
EU’s adoption of a stringent fuel economy standard that phased in over 2007–2015. It begins by following
Knittel (2011)’s approach to estimate the technological frontier firms faced during this time period, con-
cluding that firms responded to the standard by decreasing officially-reported emissions without engaging
in significant downsizing. As in Reynaert and Sallee (forthcoming), Reynaert (forthcoming) finds that the
majority of the officially reported fuel economy improvements are not reflected in real-world driving data.
To better understand these effects and evaluate impacts on consumer and producer surplus, Reynaert (forth-
coming) then develops a model of the EU auto industry, incorporating consumer preferences, marginal costs,
costs of technology adoption, and costs of gaming. The paper concludes that the EU fuel economy standard
induced firms to comply through a combination of technology adoption and gaming, with little contribution
from sales mix shifting or downsizing of other attributes. The standard reduces both consumer and producer
surplus, though by less than if firms were restricted to comply only through sales mix shifting. Relating
to Ito and Sallee (2018), Reynaert (forthcoming) also shows that the fact that the standard is weight-based
substantially curtails the CO2 emission reductions achieved by the policy, while redistributing surplus away
from French and Italian automakers and towards German automakers.
3.2.5 Lessons learned on fuel economy standards, and paths for future work
The past decade has seen tremendous research progress into firms’ responses to fuel economy standards
and the implications of these responses for consumer surplus, firms’ profits, and GHG emissions. Across
settings in Europe, Japan, and the U.S., the papers discussed above offer a useful set of tools for modeling
the variety of ways firms might respond to increasingly stringent standards: fleet-mix shifting, trading off
fuel economy against vehicles’ other attributes, increasing production costs, and gaming. Different papers
have arrived at different conclusions regarding the relative importance of these mechanisms, however. Some
of these differences might reflect genuine differences in the economic and regulatory environment across
jurisdictions studied. For instance, it may be easier for firms to game the standards in Europe than in
the U.S. But some of the differences may be related to differences in models and estimation methods, as
suggested in our discussion of Klier and Linn (2012) and Whitefoot et al. (2017). Work that reconciled these
differences and used more up-to-date data is needed to help inform fuel economy policy-making.
The existing literature has also generally not studied the extent to which fuel economy standards induce
automakers to innovate and expand their technological frontier. Such innovation is likely crucial to meet-
ing long-term decarbonization goals, and spurring such innovation is often a stated goal of fuel economy
57
policy. Research that developed models and credible evidence on fuel economy policy-induced innovation
would arguably be the single most valuable contribution IO economists could make to this area. We rec-
ognize though that such research is difficult due to challenges in measuring innovative activity, let alone its
costs and benefits. This topic may be a setting that could benefit by borrowing ideas and tools from IO re-
search on innovation in other sectors, such as Goettler and Gordon (2011)’s study of innovation in computer
microprocessor manufacturing.
3.3 Industrial organization and vehicles’ emissions of local air pollutants
In comparison to the body of papers studying consumers’ demand for fuel economy and automakers’ re-
sponses to fuel economy standards, the volume of economic research on regulation of conventional exhaust
pollutants, such as PM2.5 and NOx, is relatively modest. Two recent papers, however, have shown that ideas
and tools from industrial organization can be informative about the consequences of these regulations. Our
hope is that additional IO scholars move into this under-researched area.
First, Miravete et al. (2018) raises the possibility that environmental regulation might be motivated by
objectives other than pure reduction of emissions. Specifically, the paper considers European Union (EU)
fuel taxes and regulation of conventional pollutants (in particular, NOx) and examines how these policies
jointly serve as an effective protectionist policy in the context of international trade. The paper begins by
pointing out that EU policies have historically favored diesel vehicles over gasoline vehicles in two ways.
First, sales taxes on diesel have been substantially less than taxes on gasoline: 32 versus 46 Euro cents per
liter on average during 1991–2013. Second, EU regulation of NOx pollution is less stringent than in the
U.S., again favoring diesels because they produce substantially more NOx as a byproduct of combustion.
To evaluate the effects of these policies, Miravete et al. (2018) develops an equilibrium model of the
EU auto industry, with demand and supply specified in the spirit of Berry et al. (1995). The demand model
allows consumers’ vehicle valuations to depend both on the vehicle’s kilometers traveled per Euro (which is
also a function of fuel prices at the time of purchase) and on whether the vehicle is gasoline or diesel-powered
(to capture preferences for fuel type that are unrelated to fuel costs). The supply model allows firms’ cost
of producing a diesel vehicle to differ from the production cost of an otherwise identical gasoline-powered
vehicle.
Miravete et al. (2018) then uses its estimated model to show that the EU’s fuel taxes and emissions
regulations together induced substantial adoption of diesel vehicles by EU consumers. First, the demand
estimates indicate that consumers value fuel economy and benefit from lower prices that result from firms
not having to install NOx abatement technology on par with what was required in the U.S.. In counterfac-
tual exercises, Miravete et al. (2018) shows that either equalizing fuel taxes across gasoline and diesel or
increasing the stringency of NOx abatement standards to U.S. levels would decrease diesels’ market share
in the EU by 5–10%.
Miravete et al. (2018) closes by showing that the EU’s differentiated fuel taxes and relatively lax NOx
standards led to substantial profits for EU firms, which are specialized (relative to their counterparts in the
U.S.) in diesel technology. These results are consistent with these two policies together acting as a non-
tariff barrier to competition from foreign vehicle manufacturers. Miravete et al. (2018) estimates that these
58
policies are equivalent, in terms of effects on domestic versus foreign market shares, to a vehicle import
tariff on the order of 20%. The paper therefore highlights that environmental regulations can be used as a
tool to help protect incumbent firms, in this case to the detriment of local air quality in the EU.
The second paper we discuss also focuses on diesel vehicle emissions in the EU. Ale-Chilet et al. (2021)
studies an alleged collusive agreement between BMW, Daimler, and Volkswagen to under-comply with EU
NOx emissions standards by under-sizing their vehicles’ diesel exhaust fluid (DEF) tanks. The question
underpinning the paper is why these firms found it necessary to jointly agree to under-comply, rather than
under-comply unilaterally. There is no reason to think that consumers would value a large DEF tank, and
if anything consumers would seem likely to prefer a small tank since larger tanks reduce available trunk
space. A potential answer to the question is that the firms perceived that the expected sanction they faced
from colluding was less than that from unilaterally violating the regulation, either because the probability of
detection was lower (for instance, if the firms agreed not report one another to the regulator) or because the
expected penalty conditional on detection would be lower.
Ale-Chilet et al. (2021) quantifies the three automakers’ incentive to collude by specifying and estimat-
ing a differentiated product demand and Bertrand competition model similar to those discussed in section
3.2 above. The demand specification includes trunk space as a characteristic, and the marginal cost specifi-
cation includes DEF tank size as a characteristic. The estimates confirm that consumers value trunk space
and that DEF tank size is costly, underscoring the point that firms did not have an incentive to over-comply.
Moreover, a natural property of Bertrand competition is that the profits of any given firm are an increasing
function of the compliance of its competitors. Thus, absent differences in expected compliance penalties, it
is difficult to rationalize why the three firms felt compelled to collude rather than unilaterally under-comply.
Ale-Chilet et al. (2021) finds that collusion must have reduced the expected penalty by 188–976 million
Euros in order to rationalize the collusive agreement.
The analysis in Ale-Chilet et al. (2021) then leads to two further policy-relevant points. First, the col-
lusion led to increases in both producer and consumer surplus, in contrast to conventional price-fixing col-
lusive agreements that increase firms’ profits at the expense of consumers. However, these surplus gains
were tied to substantial increases in NOx emissions, which if evaluated using dose-response functions and
a value of a statistical life from the literature, outweigh the collusion’s private surplus benefits. Second, the
paper highlights that in situations where environmental regulation is weak—in this case, the regulator did
not monitor on-road emissions, and firms were unable or had agreed not to monitor and report each other—
antitrust enforcement can play a role in buttressing enforcement. In this case, the collusion was detected as
a consequence of a European Commission investigation triggered by the Volkswagen emissions scandal in
the U.S..
3.4 Consumers’ fuel search behavior
In this section, we consider consumer search for gasoline stations. We devote time to this topic for a
number of reasons. First, a pressing environmental policy question concerns how vehicle markets will
transition away from gasoline fuel towards alternative fuels, and especially towards electric vehicles (EVs).
As highlighted in particular by Dorsey et al. (2021), studying how consumers make fueling decisions today
59
can help inform what kinds of EV charging station densities and business models for EV charging are likely
to be successful. Second, this literature has successfully drawn a close connection between consumers’
search behavior and retail markups charged by gasoline stations. These findings are substantially important
on their own given the sheer size of this industry, but they are likely also relevant to other settings (for
instance, grocery staples) in which prices change frequently and consumers must undertake some costly
search to learn about firms’ prices.
We begin by discussing the contributions from Chandra and Tappata (2011), which presents a simple
model of consumers’ search behavior and uses daily station-level price data to test it. The paper begins
by pointing out the importance of considering both consumers’ search problem and firms’ pricing problem
when formulating theoretical predictions that one wants to subsequently test. In particular, if consumers
undertake costly search in a rational manner, and if firms set markups rationally in response to one another
and to consumers’ searching, then the sign of the empirical correlation between equilibrium search activity
and price dispersion is not clearly predicted from theory.
To be more precise, consider the following simplified version of the model presented in Chandra and
Tappata (2011), which is tailored for application to retail gasoline markets.24 Consumers have unit demand
for fuel up to their valuation v. A share λ of consumers are “shoppers” with zero search cost, who will
always buy gasoline from the lowest-price seller. The remaining consumers 1 − λ each have a search cost
s drawn from a distribution G(s) with bounded, positive support. Paying this search cost reveals all market
prices to the consumer (thus, this model best maps to a situation in which consumers use a price information
website rather than driving from station to station). Let µ ≥ λ denote the overall share of fully-informed
consumers, consisting of both the shoppers and those who choose to undertake costly search.
Firms in Chandra and Tappata (2011) all have an identical marginal cost c ≤ v. The equilibrium
conditions are then that each firm’s price p must be a best response to all other firms’ prices, given µ, and µ
must reflect optimal search given firms’ prices.
Following Varian (1980), given a search intensity µ the Nash equilibrium for firms’ prices will involve
mixed strategies, with firms drawing prices from a distribution F (p) with support on [p∗(c, v, µ), v]. The
lower bound p∗(c, v, µ) ≥ c exists because firms always have the alternative of setting the monopoly price
v and earning profits from the share 1−µ of consumers who do not search and are evenly distributed across
firms. p∗(c, v, µ) is increasing in marginal cost c and decreasing in search intensity µ. Equilibrium price
dispersion, measured as the standard deviation of p, is non-monotonic (specifically, reverse U-shaped) in µ:
at µ = 0 (no consumers search) all firms set p = v, and at µ = 1 (all consumers search) all firms set p = c.
But for µ ∈ (0, 1) price dispersion is non-zero.
On the demand side, consumers’ marginal benefit from searching increases with price dispersion and is
therefore also a reverse U-shaped function of search intensity. The search cost of the marginal searcher is
strictly monotonic in search intensity for µ > λ, so there is then a unique equilibrium search intensity µ∗
that equates the cost and marginal benefit of searching for the marginal searcher.
Chandra and Tappata (2011) then uses this model to make the point that the effect of a change in con-24We omit Chandra and Tappata (2011)’s comparative statics regarding the number of firms and do not present here the full
analytic functions that characterize equilibrium.
60
sumers’ search costs on equilibrium price dispersion is ambiguous. Clearly, a decrease in the cost of search-
ing will increase consumers’ incentive to search, holding price dispersion fixed. However, price dispersion
is not fixed in equilibrium, and depending on the initial search intensity, price dispersion can either increase
(if the initial search intensity is very low) or decrease (if the initial search intensity is very high) in response
to an increase in search intensity. Thus, a decrease in consumers’ search costs can potentially increase or
decrease equilibrium price dispersion, depending on the initial search intensity.
In contrast, the effect of decreased search costs on the average price level is unambiguous: average
prices decrease. The effects of changes to firms’ marginal cost c on the equilibrium are also unambiguous.
An increase in c will increase the average price level and compress the price distribution, leading to a
decrease in search intensity.
Chandra and Tappata (2011) then examines whether these predictions are borne out in data on daily
gasoline station prices, using 18 months of data from four U.S. states. The empirical work begins by pro-
viding evidence that gasoline stations appear to use mixed strategies. Chandra and Tappata (2011) organizes
the data into pairs of nearby stations and finds that “rank reversals” of which station charges the higher price
are a frequent occurence. Of course, such reversals could be caused by other factors, including differences
in the time-of-day at which stations change their prices. To isolate the mixed strategy mechanism, Chandra
and Tappata (2011) compares rank reversals at stations that share a corner versus reversals at stations that
do not. The logic behind this test is that search costs are zero for stations that share a corner, but not for
more distant stations. Thus, the search model would predict more rank reversals for distant stations than for
corner stations. This prediction is borne out in the data, supporting the idea that gasoline stations are using
mixed strategies to set prices.
Next, Chandra and Tappata (2011) tests the comparative static that price dispersion should decrease with
firms’ marginal cost. Using wholesale gasoline prices as the measure of marginal cost, it finds support for
this result using both a pooled regression of market-level dispersion on marginal cost, and a regression that
isolates within-market variation over time using market fixed effects.
The empirical results in Chandra and Tappata (2011) therefore find support for the predictions implied by
a consumer search model. Its tests are indirect though, in the sense that the paper examines equilibrium price
dispersion but does not observe search behavior itself. Lewis and Marvel (2011) and Byrne and de Roos
(2017) therefore build on Chandra and Tappata (2011) by studying data that directly measure consumers’
search activity on online price platforms.
Lewis and Marvel (2011) examines data from GasBuddy.com, a U.S. website that allows users to find
station-level gasoline prices. Using data on the website’s traffic, the core empirical finding in Lewis and
Marvel (2011) is that search increases when overall gasoline prices are rising (e.g., due to an increase in
wholesale fuel prices) and decreases when gasoline prices are falling. Lewis and Marvel (2011) then further
shows that gasoline stations’ respond to these search tendencies in a way that is consistent with the model
in Chandra and Tappata (2011). When prices are rising and search activity increases, retail price dispersion
and margins decrease. The opposite occurs when prices are falling.
Thus, the firm-level behavior documented in Lewis and Marvel (2011) is consistent with a standard
rational search model, but consumer behavior is not. In the Chandra and Tappata (2011) model, consumers
61
should search less, not more when wholesale gasoline prices are rising and price dispersion is falling. The
findings in Lewis and Marvel (2011) therefore suggest that consumers’ search behavior is influenced by
factors such as salience that are not included in standard search models like that used in Chandra and Tappata
(2011).
Byrne and de Roos (2017) studies consumers’ gasoline search behavior in Perth, Australia. This market
is characterized by regular “Edgeworth cycles” (Maskin and Tirole, 1988), wherein each week stations
increase their prices substantially on Thursday, gradually undercut one another over the following days, and
then jump up again on the following Thursday. Using data from the “Fuelwatch” price discovery site, Byrne
and de Roos (2017) finds three facts. First, search activity peaks on Wednesdays, when prices are at their
lowest. This fact is consistent with a form of intertemporal search in which consumers are aware that prices
will spike the following day. Second, search activity is higher than average on Thursdays (though not as high
as on Wednesdays) when prices are at their highest. This result is consistent with rational search models,
since price dispersion is large on Thursdays because not all stations initially jump to the same price level.
Third, and finally, even conditioned on day of week effects, search levels are higher when price dispersion
is higher. The paper does not pin down the precise mechanism behind this result, but it is consistent with the
prediction from the Chandra and Tappata (2011) model that changes in firms’ costs will lead to a positive
correlation between price dispersion and search.
Finally, Dorsey et al. (2021) connects consumers’ fuel search behavior to its potential implications for
EVs. As we discuss at greater length in section 3.5 below, an important driver for consumers’ EV demand is
likely to be the availability of EV charging infrastructure. Dorsey et al. (2021) examines one mechanism by
which increased EV charger density might influence preferences for EVs: decreased time needed to search
for or drive to a charging station. The paper is especially unique because it is the only paper that we are
aware of that studies drivers’ on-road (rather than online) fueling station search behavior.
Dorsey et al. (2021) takes advantage of a unique setting in Michigan in which the second-by-second
behavior of 108 drivers was tracked over six weeks.25 The data enable the paper to identify refueling stops,
and the paper matches these stops to station-level retail gasoline price data.
The empirical exercise at the center of Dorsey et al. (2021) is an analysis of how drivers decide where to
refuel. The idea is that drivers often face a trade-off between choosing a station that does not require extra
time to reach (i.e., it is directly on the drivers’ route) but has a high price, versus a station that is off-route but
has a low price. One new descriptive fact that comes from the paper is that drivers typically do not deviate
far from their route to buy gas, and nearly 50% of refuelling stops do not involve any deviation at all.
A seemingly natural way to model drivers’ decision would be a discrete choice model in which the utility
driver i receives from stopping at station j on day t is a function of the price pjt and travel time tijt:
Uijt = −αpjt − γtijt + εijt (20)
The model in equation (20) assumes that drivers are fully informed about prices at other stations. Be-25The data on the 108 drivers studied in Dorsey et al. (2021) were originally collected as part of an engineering study of crash-
warning technology. All 108 drivers drove identical cars during the study period, and they had to purchase fuel using their ownfunds.
62
cause that assumption is unlikely to hold, Dorsey et al. (2021) instead assumes that consumers respond to a
perceived price pjt that is a weighted average of the real time price pjt and the average price pj at station
j. This model nests the full information model from equation (20) if the weight on the real time price is
1. This weight is identified in the model by the extent to which drivers’ respond to fluctuations in stations’
prices over time versus variation in average prices across stations.
The estimates in Dorsey et al. (2021) yield several useful results. First, the estimated price weights
indicate that drivers respond substantially more to average price variation across stations than to real time
prices. Second, the estimated ratio γ/α represents drivers’ value of time while searching for fueling sta-
tions.26 This estimated value of time is about $25 per hour, which is more than double the value of time that
would come from a conventional estimate based on one-half of average gross wages (Small, 2012). This
estimate—which is closely related to the initial descriptive fact that drivers tend not to go far out of the way
to buy cheap gasoline—implies that consumers experience a strong disutility from spending extra time on
the road.
Dorsey et al. (2021) closes by noting that this disamenity from extending the duration of a trip poses
a challenge to EV adoption: drivers will demand a dense EV charger network. The paper makes this
point precise using a simple, but illustrative, Salop (1979) spatial competition model, in which the surplus-
maximizing number of charging stations depends on stations’ fixed cost, the number of EVs, and the travel
time cost. In this model, using the Dorsey et al. (2021) value of time estimate rather than the conventional
one implies a roughly 50% increase in the number of charging stations needed to maximize social surplus.
We see Dorsey et al. (2021) as an important first step in using observations of drivers’ current fueling
behavior to draw implications about future EV and EV charging markets. If anything, we suspect that Dorsey
et al. (2021) likely underestimates the overall charging-related challenge to EV adoption, since at least
with current technology EV charging requires substantially more refuelling time than does a conventional
gasoline station. Future research that further probed the extent to which drivers value minimizing time
spent searching for fuel—and time spent fueling once fuel is found—would be valuable in informing future
business models for EV charging, including whether charging will be dominated by home and work chargers
or whether independent chargers are a viable business model.
3.5 Markets for EVs and EV charging stations
Recent decreases in the production costs of EVs, combined with improvements in EVs’ quality and battery
range, suggest that EVs may transition from a niche product to the mainstream by the end of this decade. A
key challenge, however, to substantially increasing EV penetration is that EVs require the development of a
charging network. Because the value of an EV is increasing in the availability of charging, and because the
value of a charging station is increasing in the number of circulating EVs, EVs and charging stations should
be thought of as indirect network goods. The presence of network externalities then raises at least two ques-
tions that have been examined in recent work: (1) given many jurisdictions’ policy objective to increase EV
penetration, to what extent should governments subsidize EVs versus subsidize charging stations; and (2)
what are the economics of charging standards, and should different manufacturers be required to adhere to26Dorsey et al. (2021) also uses information on the average amount of fuel purchased per stop to arrive at a value of time estimate.
63
an interoperability standard? These questions are also closely related to broader questions about the eco-
nomics of network goods in a variety of industries, and whether governments should enforce interoperability
standards between different service providers. The models and estimations strategies used in these papers
could potentially serve as jumping-off points for work studying indirect network effects in other settings.
3.5.1 Indirect network effects and EV incentive policies
In EV markets, indirect network effects arise because consumers’ demand for EVs depends on the availabil-
ity of charging stations, and firms’ incentive to invest in charging stations depends on the number of EVs
in circulation. For a government that seeks to increase EV adoption via subsidies, a question then arises of
whether subsidies are better spent on EVs themselves, on charging stations, or on some of both.
This question is addressed in Li et al. (2017) and Springel (forthcoming). Li et al. (2017) studies the
EV market in the U.S. from 2011–2013, and Springel (forthcoming) studies the EV market in Norway from
2010–2015. The models studied in these two papers are similar, and they are used to both contribute to our
understanding of EV incentives while also providing a research design for estimating and simulating models
with indirect network effects. The policy-relevant bottom lines of these papers are also remarkably similar:
at the overall subsidy levels in force during the periods studied, subsidies for charging stations were roughly
twice as effective at inducing EV adoption than were subsidies for EVs themselves.
We focus our discussion on Springel (forthcoming), given the relative maturity of Norway’s EV market.
Norway arguably has the most aggressive EV subsidization policy in the world, and EVs have recently
accounted for the majority of new vehicle sales there. Norway subsidizes EVs by exempting them from the
country’s large registration and value-added taxes, and it also provides financial support for the installation
of charging stations.
Li et al. (2017) and Springel (forthcoming) ask how counterfactual subsidies that change the amount of
government expenditures on EVs versus charging stations would affect EV adoption rates. Answering this
question is challenging because it requires understanding how consumers value both charging networks and
EVs themselves. Springel (forthcoming) begins by specifying a demand model that incorporates network
externalities into EV buyers’ utility functions. Specifically, demand comes from a random utility discrete
choice model, where the value each consumer i obtains from vehicle model j in market m is given by:
uijm = βNi logNjm − αipjm + βki xkjm + ξjm + εijm (21)
As usual, equation (21) includes the vehicle price pjm, a k-dimensional vector of observed vehicle
characteristics xkjm, an unobserved quality term ξjm, and an idiosyncratic model preference εijm. The novel
term in this model is Njm, which (if model j is an EV) denotes the number of available charging stations
in market m.27 βNi then denotes consumer i’s preference for charging infrastructure, and the log functional
form builds in a declining marginal utility of charging network density.
The inclusion of Njm in the demand model leads to an identification challenge, on top of the usual
challenge of price endogeneity: if consumers in marketm have an especially strong preference for EVs, that27In Springel (forthcoming), all EVs can charge at any charging station.
64
will lead firms to install more chargers in that market, leading to upward-biased estimates of βNi . Springel
(forthcoming) addresses this problem by using charging station subsidies, which vary substantially over
time, as an instrument for Njm.28 Ideally, the functional form by which Njm enters equation (21) would
allow for multiple parameters, so that the level and decline of marginal utility of station density could vary
independently, but credibly estimating such a model was likely infeasible given limited instruments and
data.
On the supply side, Springel (forthcoming) focuses on the incentives of charging station owners. Vehicle
manufacturers are not explicitly modeled, so vehicle supply is effectively assumed to be perfectly elastic
(and all EV subsidies therefore fully pass through to consumers). Because Springel (forthcoming) does
not observe charging prices and markups, the paper specifies a reduced-form charging station entry model
in which logNjm is an affine function of logQjm, the total number of electric vehicles registered in the
market, along with controls for charging station incentives. As with the demand model, the presence of
network externalities creates an identification challenge, and the paper therefore instruments for logQjm
using the density of gasoline fueling stations.29 The logic of this instrument is that fueling stations affect
EV sales by affecting the relative fuel cost of gasoline vehicles versus EVs.
Both the demand and supply estimates in Springel (forthcoming) are consistent with strong network
effects, which then lead to a series of positive and normative implications. First, an interesting result is that
when network effects are accounted for, the cross-price elasticities between EV models are often negative,
rather than positive as one would normally expect for substitute goods. This complementarity arises through
feedback effects that arise from charging station entry.
Second, the sensitivity of EV demand to charging station density is sufficiently strong that charging
station subsidies are actually twice as effective, per Norwegian kroner spent and at current funding levels,
at inducing EV adoption than EV subsidies.30 However, the paper also finds diminishing returns to subsi-
dizing charging stations, so that the gains from additional subsidies to charging stations would not persist
indefinitely.
3.5.2 Compatibility between charging networks
Another question for EV deployment concerns the compatibility of charging networks, especially for high-
speed (level 3) chargers that can fully charge an EV battery in 30 minutes. Currently, different EV manu-
facturers have developed at least three distinct, incompatible standards for this charging technology. As a
consequence, a Nissan Leaf EV cannot, for example, take advantage of a level 3 charging station designed
for a Tesla.
Should the government mandate interoperability? Li (2019) presents a model showing that, despite
the obvious appeal to EV consumers of being able to charge at any level 3 station, there is a trade-off28For its U.S. context, Li et al. (2017) uses a Bartik-style instrument: an interaction between time-series variation in national
public charger investment with spatial variation in the number of grocery stores (which are common sites for public EV chargers).29Li et al. (2017) uses lagged gasoline prices as the instrumental variable for the stock of EVs when estimating its charging
station supply equation.30Li et al. (2017)’s intuition for its similar result is that early adopters of EVs are not especially price-sensitive, and they instead
greatly value the ability to easily charge their vehicles.
65
here because interoperability requirements will depress EV manufacturers’ incentives to invest in charging
networks.
In Li (2019)’s model, EV manufacturers act as Bertrand oligopolists. These firms, in addition to the usual
vehicle pricing decision, must also decide how much to invest in their level 3 charging network. Consumer
demand is similar to that given in equation (21) from Springel (forthcoming), but in Li (2019) demand
depends on both the overall density of level 1 and 2 chargers (which can charge any EV) and the number of
level 3 chargers associated with the specific brand.31 As in Springel (forthcoming), charging station density
enters via a log functional form so that there are diminishing returns to adding additional chargers to the
network. The fact that demand for a particular EV model j depends on level 3 chargers installed only by the
manufacturer of j gives that manufacturer an incentive to install chargers. In contrast, in a model in which
interoperability mandates mean that EV owners can use any level 3 charger, manufacturers have an incentive
to free-ride on other manufacturers’ installations, and the total number of chargers is sub-optimal.
Li (2019) estimates its model by taking advantage of variation in charging station build-out caused by the
2009 American Recovery and Reinvestment Act, which allocated funds for EV chargers across counties in
ways that the paper argues are plausibly orthogonal to EV demand. In addition, identification of consumers’
price sensitivity makes use of both the standard Berry et al. (1995) instruments and variation in federal and
state EV subsidies.
The main counterfactual in Li (2019) then studies market outcomes under an interoperability standard.
Despite finding that such a standard would decrease charging station investment by about 3%, the benefits to
consumers from interoperability would increase EV sales by 21%. Thus, the paper concludes that the direct
benefits from an interoperability standard outweigh its impact on firms’ incentive to reduce charging station
investment.
3.5.3 Paths for future research on EVs and EV charging
Li et al. (2017), Springel (forthcoming), and Li (2019) together highlight the importance of indirect network
effects between EVs and EV charging stations for driving EV adoption. These papers represent important
first steps, but much more is needed to understand the rapidly growing market for EVs. Especially as EV
market penetration increases in the coming years, we see promise in the following research topics:
• Fuel prices and EV adoption. How does variation in the relative prices of electricity and gasoline
affect EV adoption? Answers to this question will inform policies such as gasoline taxes or subsidies
for purchases of electricity provided through EV chargers.
• EV charging networks. To what extent will EV drivers’ charging needs be met by home charg-
ing, workplace charging, or third party charging stations? How might drivers substitute across these
options? Will competition across workplace and third party charging stations resemble current mod-
els for retail gasoline station competition, or will the organization of this industry be substantially
different?31Li (2019)’s model of how demand depends on the charging network is spatially rich, as it accounts both for the density of the
charging network around the consumer’s home as well as the ability of the network to facilitate inter-city travel.
66
• Innovation. How do EV policies affect innovation in EV technology, especially battery capacity
and charging time? How are EV innovation incentives affected by oligopolistic competition among
automakers?
4 Electricity markets
Electricity is an input to virtually any economic activity. Electricity is also a unique commodity when it
comes to its economic characteristics. While it is a homogeneous good, it is also, differently than in other
industries, concurrently produced with a wide range of technologies (e.g., coal, natural gas, hydro power,
wind and solar, etc.). Among other unique features, electricity has been extraordinarily difficult to store
economically (and still is today, although this is changing). The difficulty of storing electricity, coupled
with generally quite inelastic demand in the short- and medium-run, leads to substantial price fluctuations
and seasonality, as well as moments of scarcity in which market power can be substantial. The presence
of capacity constraints and difficulties in storage is in fact behind the co-existence of such a wide range of
generation technologies, depending on their level of utilization and flexibility. The delivery of electricity is
also distinct and mainly done via the electric grid, whose operation needs to be centralized and coordinated.
The process of coordinating the second-by-second match between demand and supply while ensur-
ing that the electricity grid can deliver power is done by centralized authorities. Electricity markets have
emerged in the last 30 years to help make this process more transparent and cost effective, leading to a large
body of research aimed at understanding the efficiency implications of these changes, as well as the poten-
tial pitfalls, such as an enhanced ability for firms to exercise market power. These are core questions at the
heart of the study of industrial organization. The fact that these markets were traditionally regulated (and
some of them (or parts of them) still are) also makes it particularly suited to study canonical questions in the
economics of regulation literature. Additionally, the existence of highly granular data with plausible shifters
both on the demand and the supply side make electricity markets particularly attractive for IO economists.
Electricity markets are currently again in the midst of enormous transformation, with a rapid decar-
bonization of electricity generation and an acceleration of the electrification of many major activities, such
as transportation. Efficiently divesting from a carbon-intensive to a carbon-free grid will be essential to min-
imize the costs of the energy transition. Therefore, the analysis of these markets can have crucial impacts to
the overall economy as well as the environment and our future prospects. The field of industrial organization
can contribute with its tools to analyze market design and market structure to ensure a more efficient and
equitable transition.
It would be impossible to cover all of the papers relevant to the study of electricity markets. This section
focuses on some of the papers that are most relevant to IO economists while trying to highlight unifying
themes across them.
4.1 The restructuring of electricity markets
Since the 1990s, electricity markets have undergone large transformations in market organization with the
emergence of liberalized electricity markets in many countries of the world. Traditionally, electricity mar-
67
kets had been operated under rate-of-return regulation and often in the form of a vertically-integrated natural
monopoly in all of its segments: generation, transmission and distribution, and retailing. However, several
markets transitioned away from this regulated vertical structure and opened up the generation and retail
segments of the electricity sector to competition. This tremendous shock to how electricity markets are or-
ganized has been a source of a large amount of IO research studying the performance of wholesale and retail
markets under imperfect competition, optimal market design, and restructuring’s impacts on productivity
and efficiency, among other topics.
In the United States, the process of restructuring in wholesale electricity markets has been tumultuous
and slow, which has also been an interesting source of variation to researchers. The restructuring efforts
started in markets such as California, New England, the Pennsylvania-Jersey-Maryland market (PJM), but
the restructuring of other markets was put on hold after the market crash of the California electricity markets
in the 2000s, leading to varied forms of market structure and competition. Since then, several regions have
transitioned into formal organized markets, but large areas still remain under cost-of-service regulation for
the generation segment.
Borenstein and Bushnell (2015) provides a retrospective of the liberalization process in the United
States, highlighting its successes and failures. Bushnell et al. (2017) reviews the findings of the literature in
more detail than what we can provide here.32
4.1.1 Aggregate impacts of restructuring
From an industrial organization perspective, the halt of the restructuring process in the United States pro-
vides a unique opportunity to study complex markets that are quite homogeneous in their main purpose (the
provision of electricity through the grid) but are organized with a wide range of regulatory tools. A literature
has emerged that treats the presence of these organizational forms as a quasi-natural experiment, comparing
markets before and after restructuring to similar ones that did not undergo this transformation.33 Even if it is
difficult to ensure a purely apples-to-apples comparison between heterogeneous regions, this large variation
in regulatory form is distinct relative to many other markets and worth studying.
Cicala (2020) evaluates the overall impacts of the transition to market-based wholesale electricity oper-
ations using the staggered transition to markets from 1999-2012. It assembles a detailed dataset of hourly
operations of virtually all power control areas (PCAs) in the United States. The data include hourly demand,
hourly generation of a large share of power plants, and engineering estimates of the costs of operating power
plants. The paper presents a decomposition of the hourly costs of electricity generation at the PCA level. It
compares the actual costs, calculated with the generation and engineering cost data, to the costs that result
from minimizing production costs, holding the observed quantity fixed. This is an easy calculation because,
by holding quantity fixed, one does not need to worry about how energy flows between PCAs. Additionally,
it computes the surplus that arises from trade by comparing the additional opportunity costs of produc-
ing at autarkic quantities when compared the observed ones. Because the counterfactual costs of autarkic32See also Kwoka (2008) for an earlier assessment and review of the impacts of restructuring in the US electricity sector.33It is important to note that this regulatory transition is not unique to the United States. Other papers have used a similar diff-in-
diff strategy to study the impacts of increased restructuring in other regions. See for example Malik et al. (2015), which finds thatrestructuring in India had limited impacts on productivity of state-owned plants.
68
quantities are not observed, the paper’s method uses the idealized curves instead.
The paper compares these two measures of performance before and after restructuring using a differences-
in-differences strategy. It finds that the electricity markets that transitioned to centralized markets had large
gains from trade when compared to other regions, consistent with previous papers that had documented
large increases in trade after the restructuring of their operations (Mansur and White, 2012). The regions
that joined market-based wholesale electricity markets benefited from an increase in gains from trade of
55%. Departures from the idealized merit order curve within a market decrease by 16%.
4.1.2 Plant-level evidence of the impacts of restructuring
Cicala (2020) offers a comprehensive data set of all markets in the United States, but it does not quantify the
mechanism behind the improved efficiency, other than the increases in trade. Other papers have explored
the mechanisms more directly using plant-level data. Fabrizio et al. (2007) tests the use of labor and other
inputs into the production process by using data from utility-owned power plants using a production function
approach that models energy inputs as Leontief to motivate their reduced-form regression analysis. The
paper finds evidence that plants that are expected to be restructured reduce their labor expenses by 3% and
their non-fuel expenses by 5-10%. One limitation, however, is that the plant-level information is limited
once power plants are restructured, as the regulation no longer requires the reporting of labor and other
materials.
Davis and Wolfram (2012) uses a similar diff-in-diff strategy to quantify the efficiency impacts from
restructuring in the operation of nuclear power plants. It documents significant increases in the utilization
of nuclear power plants after deregulation, with a central estimate of 10% increased operating performance,
which does not seem to come at the expense of decreased safety (Hausman, 2014).
Whereas previous papers examine the impacts of regulation from an ex-post econometric point of view,
these papers do not provide structural modelling that can help quantify the impacts of counterfactual reg-
ulatory regimes. In light of all the changes that are affecting electricity markets, we aspect the differing
performance between regulated and restructured markets to reemerge as an important area of study. IO
models of investment under different regulatory benchmarks could help understand the incentives provided
by different regulatory regimes. The work in Eisenberg (2020) examining the impacts of regulatory wedges
in the performance of the Chinese electricity market is a great step in that direction.
4.1.3 Natural monopoly regulation in distribution
While the generation and retail sector have been restructured in several markets, the distribution of electricity
remains a natural monopoly. Therefore, it is subject to the challenges of providing incentives for efficient
operations and the provision of quality while regulating prices.
Lim and Yurukoglu (2018) studies the provision of electric distribution services with a comprehensive
dataset of measures of quality of service and rates for most of the United States service territories. The
paper studies the political economy of the rate-setting process and the potential commitment and moral
hazard issues involved. Regulators might be tempted to change their regulation (ratcheting effect) while
distribution companies might want to under-provide effort. Regulators have two tools to manage the utility:
69
provide an incentive to capital via rate-of-return, and provide penalties for the under-provision of quality
(e.g., based on standard measures of interruptions of service, which is their data on provision of quality).
The paper shows that the regulator will favor a high rate-of-return and low penalties for the under-
provision of quality in environments in which it weighs the utility of the firm more than consumer surplus.
The paper then builds a structural model that estimates regulators’ preferences as well as the costs of quality
provision.34 In the model, preferences are proxied for using political preferences, and more conservative
territories end up favoring higher investment and lower quality. The higher investment is not necessarily
detrimental for welfare in the presence of commitment problems, but these preferences also lead to higher
costs to consumers and lower quality of service.
Lim and Yurukoglu (2018) presents a stylized model with limited heterogeneity, only via the political
orientation of different service territories. The monopoly model is also highly stylized. Given the need to
modernize the distribution grid and incorporate new innovations, such as distributed generation and voltage
control, understanding the microeconomic underpinning of providing incentives for an efficient transition
seems an area of important of study.
Mahadevan (2021) also studies the political economy of electricity distribution, examining India rather
than the U.S. Using administrative billing data, Mahadevan (2021) finds evidence that an electric utility
substantially under-bills regions that supported the ruling party in a recent election. This under-billing
occurs despite the fact that these regions, if anything, appear to consume more electricity than other regions,
as measured using satellite nighttime lights. The under-billing result is supported by forensic data work
revealing that bills in these regions are excessively likely to be round numbers and violate Benford’s Law.
The paper closes by estimating consumers’ demand (accounting for the fact that some of the billing data
are manipulated) and showing that the increase in consumer surplus created by the billing manipulation is
outweighed by the decrease in producer surplus (which must ultimately be covered by ratepayers or public
funds).
Finally, the distribution of natural gas, also a natural monopoly, will also undergo dramatic changes
given the need to decarbonize our economies. In particular, the provision of natural gas via the distribution
network is expected to decline, creating a stranded asset problem. More work should go into understanding
how this process will unfold, and how to properly incentivize and pay for such exit, a topic that has started
to be recently explored in Davis and Hausman (2021).
4.2 Market power in wholesale electricity markets
Several papers have examined the performance of wholesale electricity markets with a focus on the potential
exercise of market power. Before getting into the specific papers and contributions, it is useful to consider a
simple framework of optimal behavior by a strategic firm participating in the electricity market.
Consider an electricity market at time t with demand given by Dt(p). In the electricity market, several
firms j = 1, . . . , J are competing in supply curves: a schedule specifying their willingness to produce
(quantity) at different price levels, Sjt(p). One can define the residual demand, left to a given firm i, as
34For the reader interested in these topics, papers discussed in section 4.5.2, 5.2.2, and section 5.3.1 are closely related to thisstrand of the literature.
70
the demand left to that firm at a price p, after taking into account the willingness to produce by the other
participants, i.e., RDit(p) = Dt(p)−∑
j /∈i Sjt(p). It is crucial to consider the impact of the other firms on
the demand left to a given firm, as otherwise the demand function tends to be inelastic, leading to unrealistic
predictions about the incentives to exercise market power.
Notice that the presence of a residual demand curve is not unique to electricity markets. One could
envision that firms have an implicit supply function of their willingness to supply at different prices for
other goods. The key in the electricity market context is that firms explicitly compete in this way in the
market. Furthermore, such curves are often observed in the data, and therefore one can be much more
explicit about the presence of such objects. In the absence of data on supply curves, bounding assumptions
can be made, typically ranging from Bertrand competition to Cournot as suggested by Supply Function
Equilibria (SFE) models (Klemperer and Meyer, 1989).
Once the residual demand for a given firm has been defined, the individual problem of the firm, taking
the behavior of others as given, can be stated as,
maxSi(p)
E[p×RDit(p, ε)− C
(Si(p)
) ], s.t. Si(p) ≤ Si(p′) for p < p′, (22)
where the introduction of ε reflects that the firm might be uncertain about the exact shape or level of the
residual demand, and the constraint states that firms need to be willing to supply weakly more quantity at
higher prices, as often enforced in electricity auctions.
Abstracting away for now from uncertainty and monotonicity constraints, the program in (22) becomes
equivalent to the firm choosing the optimal price to set along the residual demand curve that maximizes net
revenue.35 Assuming that the shape of the residual demand satisfies regularity conditions, the optimization
problem leads to the following first-order condition for a profit-maximizing firm:
p = C ′ − RDit(p)
RD′it(p). (23)
This expression is equivalent to the usual Lerner condition in introductory monopoly pricing, p−C′
p = − 1η ,
where η is the elasticity of the residual demand in this case: the more elastic the residual demand, the lower
the markup. The expression also highlights that, holding the slope of residual demand fixed, firms will
tend to exercise more market power when they are producing larger quantities; i.e., a high RDit, which in
equilibrium equals the quantity produced by the firm, Qit.
4.2.1 Estimating market power
Expression (23) can be easily brought to the data, aided by the quality and availability of cost data in
electricity markets. As a consequence, there has been a flourishing IO literature examining markups in the
context of electricity markets as well as extending this basic framework.
Wolfram (1999) empirically examines competition in electricity markets in the context of the British
electricity spot market, which Green and Newbery (1992) had numerically explored earlier in ex-ante sim-35A more explicit derivation based on the bidding strategy of firms will be covered below.
71
ulations using a similar framework. Wolfram (1999) tests conduct parameters using the above derived
first-order condition by directly constructing an analogue to the Lerner index. By using data on prices and
marginal costs, she can construct the right-hand side of the equation. By making assumption on the elastic-
ity of demand, one can then test if markups are consistent with the strategic model assumed in (22), which
would imply that markups times the elasticity (in absolute terms) should be equal to one. Because Wolfram
does not observe the strategies of the firms (or their individual output), she bounds the elasticity of demand
using results in the SFE literature (Klemperer and Meyer, 1989; Green and Newbery, 1992). Results sug-
gest that firms are far from exercising market power a la Cournot (fixed quantity strategies) and also for SFE
models that imply more competition. When compared to simulations (Green and Newbery, 1992), the esti-
mated markups also fall below those predicted ex-ante. Markups are estimated to be between 20 and 25%
during the period of study. She conjectures that such “low” markups–relative to the potential for market
power–may be explained by contracting forward positions limiting the inframarginal quantity (see below),
regulatory threats, or the threat of new entry.
Several papers have estimated markups using data from other electricity markets. Borenstein et al.
(2002) (BBW, henceforth) use data from the California electricity market to examine the crisis suffered
in 2000-2001, during which prices skyrocketed. The identification strategy in BBW is to consider market
power as the residual explanatory factor behind the increase of electricity prices. There were several shocks
affecting the California market during the period of study (1998-2000), such as a severe drought, increased
NOx prices, and increased natural gas costs. Therefore, it is important to account for such cost shifters.
BBW calculate markups from market power by comparing observed prices to the prices that would be
consistent with a first-best scenario in which firms offer their output at their marginal cost and water is
allocated efficiently. They use Monte Carlo methods to simulate potential power plant unavailabilities and
allocate the limited water resources to those hours in which they are most valuable. They also estimate
the availability of imports (the imports supply curve), which are substantial in California (around 25% of
consumption). This procedure provides an estimate of P competitive that they can they use to compute the
Lerner index:
Lernert =P observedt − P competitivet
P observedt
.
They document that the Lerner index was very modest during 1998 and 1999, but increased dramatically
in 2000 as a result of the scarcity conditions in the market. A large part of this increase is explained by
increases along the supply curve, moving to areas in which market power is more concerning.
Finally, the paper presents a decomposition of the increased costs to consumers as a function of in-
creased costs, increased competitive rents, and increased market power rents. Given the convex shape of the
electricity supply curve, competitive rents can increase substantially when moving along the supply curve,
as firms with low cost still get paid the marginal price in the market. Yet though competitive rents can ex-
plain part of consumers’ cost increase, BBW find that around 59% of increased costs between the summer
of 1999 and the summer of 2000 can be attributed to market power.
Given the large increase in profits and prices documented in Borenstein et al. (2002), what can we say
about firm conduct? Were firms exercising market power according to a Cournot model during the sample?
72
Or did they coordinate or collude to sustain even larger profits? Unfortunately, the analysis Borenstein
et al. (2002) offers little explicit guidance in answering this question, as markups are estimated directly
from the data without a strategic model of competition. To inform this question, Puller (2007) estimates
conduct parameters more explicitly in the California electricity market using equation (23). To examine
this question, the paper tests whether firm behavior is consistent with Cournot pricing. To do so, Puller
estimates the markup equation taking into account the elasticity of the fringe suppliers, which is estimated
from the data using forecasted demand as an instrument, as well as the presence of capacity constraints,
which might affect marginal costs. He finds that in the years before the crisis, the model is consistent with
Cournot behavior. During June-November 2000, the model suggests firms exercise market power above the
Cournot levels, but still far from a perfectly collusive equilibrium.
The market structure in California led to large increases in prices during the 2000 and 2001, but the
crisis was not concurrent in other restructured markets, such as the New England and PJM markets. Natural
gas prices also spiked in the Eastern states, but the effects on price increases were more modest. What
are some of the factors explaining such differences? Bushnell et al. (2008) (BMS, henceforth) studies the
importance of forward positions to explain this empirical pattern. Forward positions refer to the quantity that
the firm has already committed at a pre-established price; e.g., via a physical contract, a financial contract
for differences, or a selling position due to vertical integration at regulated (sticky) retail prices. The key is
that the price of such forwards should be fixed and not depend on the day-ahead electricity market outcomes.
In such a case, the FOC in (23) becomes:
pt = C ′it −RDit(p)− θitRD′(p)it
, (24)
where θit represents the forward position of firm i at period t. Because this quantity is already committed
at a pre-established price, the effect of forward positions is to reduce the infra-marginal quantity of the firm
(RDit − θit) and, therefore, its incentives to exercise market power (Allaz and Vila, 1993). If all output is
contracted in advance at a fixed price, the firm will have limited incentives to exercise market power. If the
firm is a net buyer in the market (RDit < θit), the firm might even be willing to exercise monopsony power.
BMS puts together data from three markets (California, PJM, and New England) in which forward con-
tracts are either observed via vertical agreements or zero (as in California). BMS then tests the predictions
from three models: a competitive model a la BBW, a Cournot model that ignores the presence of forwards,
and a Cournot model that includes forwards in the first-order condition. Its main finding is that a model
with forward contracting can rationalize quite well the observed patterns in the data. The competitive one
under-predicts prices quite systematically and the Cournot one without vertical agreements leads to an over-
statement of market power.36 The paper is useful to highlight that the absence of forward contracting can in
part explain the California crisis.36The Cournot assumption contributes in part to the massive market power predicted in the absence of forward contracts, as the
strategic firms are assumed to submit inelastic supply curves in the model (quantities), and therefore do not contribute to makingthe residual demand more elastic.
73
Figure 2: An example of supply and residual demand draws observed in electricity markets
Note: Electricity markets provide a unique setting in which the supply curve and the residual demand that firms
face is observed. Source: Authors’ construction based on data from the Spanish electricity market. See Reguant
(2014).
4.2.2 Taking electricity auctions to heart
The papers we have discussed thus far are not very explicit about the bidding process in the electricity mar-
kets they study and do not use bidding data directly. However, electricity markets are typically formally
organized as multi-unit uniform auctions. The bidding process and the strategy space in electricity markets
are well known and often observed in the data. Indeed, these rich data are a unique feature of electricity mar-
kets: the researchers get to observe the supply curves of firms, not just their realized supply in equilibrium,
thanks to detailed electricity auctions data. Going back to the maximization problem of the firm in (22),
the availability of auction data implies that the ex-post residual demand faced by the firm can be observed.
Figure 2 shows an example of bidding data of one firm as well as the residual demand that the firm faces.
Early papers in the literature such as Wolfram (1998) explore how to model the firm’s first-order con-
dition as a multi-unit auction, by representing their maximization problem as a function of the distribution
of opponent’s bids, in a private value framework. The key to understand bidding behavior is to model the
trade-off between increasing the price for all accepted units if a bid is accepted, due to the uniform format
of the auction, and reducing the probability of having that offer accepted. In equilibrium, firms submit bids
with markups above cost that are consistent with the trade-off between these two forces.
Hortacsu and Puller (2008) (HP, henceforth) shows that the first-order conditions for each bid implied
by the multi-unit auction framework can be analogous to expression (24). In particular, if uncertainty only
leads to parallel shifts of the residual demand, so that the slope at a given price p remains constant, it is
74
sufficient to know the slope of the residual demand curve at each price to determine the ex-ante optimal
supply curve.37 These ex-ante optimal bids will also be ex-post optimal under such conditions. For the
purposes of this Handbook, and given the close link to equation (24) in uniform multi-unit auctions, we will
focus on the additional value of bidding data when studying these markets rather than on the fine details
between alternative auction models.38
How does the presence of bidding data aid the identification of fundamentals in this setting? In electricity
markets, the three main potentially unknown factors are typically power plants’ costs, the forward position
of the firm, and firm behavior (i.e., model of competition). In the traditional IO literature, oftentimes firm
behavior needs to be assumed as a source of identification, but this is not necessarily the case in the electricity
context, in which firms’ supply curves and costs are often observed. Electricity auctions papers differ
in which fundamentals are taken from the data, which assumptions are imposed, and which objects are
estimated.
Wolak (2000) and Wolak (2003) use a framework of optimal bidding under the assumption of profit
maximization to estimate costs. He proposes to smooth out the residual demand curve in order to estimate
its slope, RD′it(p), via Kernel smoothing techniques. In his applications, forward contracts are observed for
one of the firms, and a rule-of thumb assumption is used for others.
The Kernel representation of the residual demand allows for taking the derivative of the residual de-
mand with respect to a power plant’s bid in a simpler fashion than if the lumpiness of step functions where
to be considered.39 In the context of electricity market auctions, in which steps are typically small and the
uncertainty around prices is relatively narrow, the probability of setting the price can be difficult to quan-
tify. Kernel smoothing is a practical way of circumventing both of these computational problems. One
can express the first-order condition as a function of the bid, where the bid has an indirect impact on the
equilibrium price p(b).
With knowledge of the forward position and under expected profit maximization, the firm’s first-order
condition for a given bid accounting for uncertainty becomes:
∑d
((pd − C ′)
∂RDd
∂pd+RDd − θit
)∂pd∂bijkt
= 0, (25)
where d represents a given draw in the empirical distribution of the residual demand faced by the firm and
bijkt is the offer made for a step k of plant j. One can estimate the model via GMM (e.g., Wolak (2007)).
The equations can also be estimated via a weighted regression approach, in which the weights are given by
the a measure of how close the bid is to setting the price (Reguant, 2014). In the limit, when one considers
the impact of the bid only when the bid is exactly marginal (b = p), this is equivalent to estimating the
first-order conditions in (24) for those bids that are exactly marginal.
Hortacsu and Puller (2008) (HP, henceforth) highlights that, in the presence of bidding data and observed37Brown and Eckert (2021) shows the required conditions for this result to hold need to be stronger if one wants to ensure that
the supply curve is weakly increasing, a common restriction in these markets.38The interested reader can refer to the auctions chapter in this Handbook for a more detailed derivation of the first-order condi-
tions with bidding data.39See the Handbook chapter on auctions for a detailed treatment of the step-function nature of supply and demand curves in
multi-unit auctions.
75
costs, one can infer the forward position of firms, θit, by examining the point at which the supply curve and
the cost curve cross. That is, in equation (24) the point at which p = C ′ is the point at which θit is equal to
the offered quantity. This inference allows one to test for behavioral assumptions, instead of assuming profit
maximization. Whereas this identification strategy requires firms to place no markup at the point at which
their offered quantity matches their contract, HP can still test if the behavior of firms is consistent with
profit maximization. HP shows that large firms behave closer to a profit-maximizing bidder, but small firms
submit supply bids that are too inelastic, leaving substantial profits on the table and significantly increasing
inefficiencies in the market, which are even larger than those induced by market power. Hortacsu et al.
(2019) revisits the results in HP and show that a k-level hierarchical model in which low-sophistication
(small) firms bid inelastic curves can rationalize the observed behavior.
The previous papers either take costs as given and estimate conduct and forward positions (Puller, 2007;
Hortacsu and Puller, 2008; Hortacsu et al., 2019), or observe forward contracts from data and back out costs
(Wolak, 2000, 2007). However, with bidding data it is possible to estimate both types of parameters at the
same time with minimal assumptions. Intuitively, the forward position is a parameter that is common at the
firm level, whereas marginal costs are specific to each plant. As long as more than one bid is observed by
some plants or temporal restrictions are considered, it is possible to estimate the cost structure and forward
contracts at the same time in a flexible way. Re-writing the first-order condition as a function of the optimal
bid helps emphasize the flexible identification of these objects under mild functional form assumptions:40
bijkt = cjt(qj)−RDit(b)− θitRD′(b)it
, (26)
where k indexes a given price bid for unit j. With mild assumptions on the shape of cjt and several offers
per unit j, a flexible estimate of cjt and θit can be recovered. The identification of the forward contracts is
facilitated by the presence of several units per firm, as this is a common parameter. Intuitively, one could
identify the two at the same time even with only one plant as long as there are more steps than parameters
in the cost function. Empirically, restricting the functional form of the cost function or the forward position
based on knowledge about the production function and institutional forward position allows for consistent
estimation of the parameters (Reguant, 2014).
4.2.3 FOC approach and pass-through analysis
The data and first-order conditions discussed above lend themselves very well for the study of pass-through
of costs. Compared to other applications in the pass-through literature, electricity markets feature a well-
known market structure formalized via centralized auctions, detailed high-frequency daily or hourly data, a
wide range of cost shocks at a similar level of frequency, and good knowledge of the cost structure of firms.
Fabra and Reguant (2014) uses data from the Spanish electricity market to measure the degree of pass-
through of emissions costs. It combines a reduced form analysis of the time series pass-through of emissions
costs with a structural analysis of the individual firm’s responses. In the time series analysis, and after
controlling for the endogeneity of emissions costs using the permit price as an instrument, the paper shows40Wolak (2007) derives similar first-order conditions with respect to quantities, as opposed to prices.
76
very high levels of pass-through. The firm-level analysis tests to which extent firms consider the full cost of
emissions in their bidding behavior. It uses a weighted regression estimation similar to equation (26) that
takes the following form:
bijkt = βcjt + γejτt + θ markupikjt + εijth, (27)
in which the markup has been estimated following the steps discussed above and is instrumented with de-
mand shifters such as temperature. The authors find that firm behavior is consistent with full internalization
of the costs of emissions, i.e., γ ≈ 1. However, they find that the pass-through of marginal costs is greatly
attenuated, with β significantly smaller than one. They relate the finding to measurement error, highlighting
the limits of using engineering-based marginal cost data in regression analysis. Kim (2021) extends the work
in Fabra and Reguant (2014) studying the pass-through of natural gas costs in the New England electricity
market. She shows that taking marginal costs as given can lead to the attenuation bias mentioned above,
highlighting how heterogeneity in the generation mix can contribute to additional bias.
4.2.4 Market power and dynamics
One limitation of most of the above papers is that the representation of marginal costs is relatively limited
and ignores substantial complexities of how power plants operate, such as the presence of ramping costs
from increasing or decreasing production, technical minimums that prevent a power plant from reducing its
output below certain thresholds, and startup costs. Hortacsu and Puller (2008) ignores dynamics by focusing
only on bidding behavior in the afternoon. Bushnell et al. (2008) highlights that their markups appear to be
negative at night, i.e., firms are bidding below their marginal costs, and explains that such biases are likely
driven by dynamic considerations. However, the proposed methodology does not directly address these
concerns.
Mansur (2008) points out that the approach followed by Borenstein et al. (2002) and others does not take
into account the complex dynamics involved in the production of electricity. The paper proposes a novel
method to estimate the cost of electricity generation without relying so heavily on marginal cost data. In
particular, it presents a flexible model of decision-making under regulation that predicts production choices
as a function of observables. The approach uses a dynamic optimization Bellman equation to motivate the
covariates that should be included, which are mostly a range of lags and leads of price cost margins. Looking
at data from the PJM market, Mansur (2008) shows that the generation patterns are much better explained
with this richer model. When looking at the impacts of restructuring, the paper draws to comparison. Using
static costs to compute a counterfactual implies that market imperfections resulted in considerable welfare
loss, with production costs exceeding the competitive model’s estimates by 13%–21%. However, actual
costs were only between 3% and 8% above the competitive levels when using predicted prices based on
pre-restructuring behavior based on the more flexible proposed methodology. The paper highlights the
importance of considering dynamics.
Other papers take a more structural approach. Wolak (2007) expands the estimation of marginal costs by
including ramping costs: costs that depend on output in the previous period. This is achieved by expanding
77
the functional form assumption of the marginal cost. The paper models the daily decisions of the firms
when they face ramping costs as well as minimum production levels. Reguant (2014) exploits a feature of
the electricity auction design in the Spanish electricity market, which enables firms to bid their fixed costs as
additional “complementary” bids. It shows how startup costs can be identified from such an auction design.
Using a finite horizon model in which firms look at several days ahead, it shows that a model accounting for
such costs helps better capture firm behavior.
Reguant (2014) also shows that a model accounting for dynamic costs helps correct the biases identified
in Bushnell et al. (2008). Whereas a static model predicts negative markups at night, a dynamic model
predicts prices that are below marginal costs for both a competitive and a strategic firm. Importantly, the
prices are higher when firms exercise market power (even if below marginal costs). Therefore, market
power still increases prices in all hours. When it comes to hours of high demand, a static model tends to
underestimate the amount of market power. If demand is only peaking for very few periods, fewer power
plants will be in operation, which limits competition.
4.2.5 Sequential markets and arbitrage
Whereas most of the economics literature has focused on the understanding of electricity markets as a daily
auction, electricity markets clear several markets for the same product (e.g., electricity delivered at 10 am
on May 10, 2011), in the form of sequential centralized markets. Electricity markets typically open the day
before delivery, the so-called “day-ahead market,” but also closer to delivery, e.g., several hours ahead or in
the real-time market. The combination of different horizons allows for better planning of the operations in
the market and can potentially enhance efficiency.
What are the strategic incentives in these markets? Ito and Reguant (2016) explores this question in
the Iberian electricity market. A simple theoretical model shows that the relatively inelastic demand in the
day-ahead market (often based on forecasted demand) combined with limited arbitrage tends to lead to a
price premium in the day-ahead market. In the Iberian market, this leads large firms to under-commit in
the day-ahead market, whereas renewables producers partially arbitrage the price differences by supplying
more power than forecasted in the day-ahead market.
To examine the welfare implications of this behavior, Ito and Reguant (2016) adapts the quantitative
model in Bushnell et al. (2008) to include two sequential markets. The model features Cournot competition
in two stages. In the day-ahead market, firms commit to their positions. In the real-time market, firms can
revise their offers and either buy or sell to change their equilibrium quantities once some uncertainty in the
market has been resolved. The first day-ahead market position acts as a forward position in the real-time
market, reducing the incentives to exercise market power by the firms. This can lead to a reduced price in
the real-time market as long as there is limited arbitrage in the market.41
The equilibrium model presented in Ito and Reguant (2016) can replicate well the observed price pre-
mium in the market. In counterfactual analysis, the authors consider the impacts of introducing financial41Until the introduction of financial participants (see discussion below), the assumption of limited arbitrage is reasonable in
these markets. Supply-side participants are limited by their power plant capacities. On the demand-side, the regulatory mechanismimposes that the day-ahead market plans for almost all expected demand, even if it is profitable to delay demand purchases in thepresence of a day-ahead price premium.
78
arbitrageurs in the market. They find that financial arbitrageurs contribute to an increase in consumer sur-
plus, by lowering the prices in the day-ahead market, but do not increase efficiency. The rationale for this
result is that arbitrageurs lower the day-ahead price but increase the real-time price, leading to larger inef-
ficiencies in the final settlement. The paper also complements the theoretical literature in Allaz and Vila
(1993) by quantitatively showing that, even with strategic distortions, the presence of a secondary market
successfully reduces market power concerns when compared to a single primary market. The paper con-
cludes that price convergence, e.g., due to the introduction of financial participants, should not be interpreted
as a sign of increased efficiency on its own.
The presence of a gap between the day-ahead and the real-time market persist in many electricity mar-
kets. Borenstein et al. (2008) shows that this was the case in California, in which the real-time price market
tended to be above the day-ahead market price. In California, large firms were net buyers of power, thus
leading to monopsony power and lower day-ahead prices. It posits that the perceived threat of regulatory
intervention prevented participants from arbitraging price differences away. However, over time several mar-
kets in the United States, including the California one, have been progressively introducing purely financial
(virtual) bidders that are actively encouraged to arbitrage these differences. Financial arbitrageurs take a
position in the day-ahead market at a given node in the network (long or short), and commit to undoing such
position in the real-time market.
Is the introduction of financial arbitrageurs beneficial? Birge et al. (2018) shows that the ability of
arbitrageurs to close price gaps is limited by their access to capital and transaction costs. It also highlights
that the complex structure of electricity prices and transmissions constraints opens the door to opportunities
for increasing consumers’ costs. Jha and Wolak (2020) studies the role of financial bidders in the California
electricity market and come out with a more positive interpretation of the role of financial traders. It presents
evidence that financial arbitrageurs do successfully remove persistent price differences in the market. Using
an event study methodology, it also presents event-study evidence that costs in the electricity market were
reduced thanks to this policy change. Mercadal (2021) highlights the role of financial bidders in limiting
the scope for collusion in the market. Using data from MISO and an event study combined with a structural
model of bidding, she finds that the introduction of financial bidders (or rather, the upcoming threat of
financial bidders) limited the markups that firms could sustain. Financial bidders also reduced the price gap
between markets.
4.2.6 Modeling transmission and market power
Another aspect that is often simplified in the modeling of electricity markets is the transmission electricity
grid. Theoretical analysis of the network features of electricity markets show that even models with simple
layouts can be extremely difficult to solve, whether only the output market is considered (Borenstein et al.,
2000) or also competition over transmission rights is modelled, in which firms can exhibit additional strate-
gic behavior by withholding transmission even if physically feasible (Joskow and Tirole, 2000). Indeed,
most papers above treat the electricity market as having a unique price as a simplifying assumption.
Due to the shift to nodal (geographically disaggregated) prices in the United States, and the growth in
congestion due to extreme weather and the increase in renewable power, ignoring transmission constraints
79
has become less palatable. Transmission constraints can substantially impact competition by creating iso-
lated areas in which firms face less competition.
In equation (23), transmissions constraints impact the location and slope of the residual demand.42
Wolak (2015) quantifies the incentives to exercise market power once transmission effects are accounted
for in the Alberta electricity market. He finds that the slope of residual demand is substantially more elastic
in the presence of no congestion, with the inverse elasticities (which determine the incentives to markup the
bids) being two- or three-fold larger in the hours of high demand. To quantify the benefits of transmission
due to reduced market power, he corrects observed bids with the ones based on more elastic residual de-
mands and finds competitiveness benefits of perceived higher transmission, even in a worse case scenario
counterfactual in which transmission does not actually expand.
Ryan (2021) models the impact of transmission constraints on market power in the Indian real-time
electricity market. Because the Indian electricity market only has few price regions, the analysis of trans-
missions constraints is somewhat easier than in nodal markets, which can have thousands of different price
points. Ryan documents that congestion leads to significant increases in market power, as measured by the
slope of residual demand, as in Wolak (2015). He then proposes an estimator that adapts Wolak (2003) and
Reguant (2014) to account for the probability of congestion in the maximization of expected profits. He then
constructs a Cournot structural model with transmission constraints and simulates the value of expanding
transmission in the Indian electricity market, finding that it would benefit exporting firms in regions with
ample supply but that face congestion when selling their power.
In the language of antitrust policy, transmission constraints affect the definition of the relevant market.
How to deal with more complicated markets in which there are thousands of price points in the network?
Mercadal (2021) contributes to the study of nodal markets by developing a machine learning methodology
to define relevant markets. Using data from MISO, she uses a clustering algorithm to find sets of prices that
are highly correlated. To discipline the clustering algorithm, which is otherwise unsupervised, she adds a
criterion function based on how well the implied relevant markets replicate observed prices. Whereas this
approach makes simplifications about the strategic nature of congestion, it is a useful way of making the
analysis of nodal markets more tractable and opens the door to analyzing a variety of important question in
these markets. We expect that machine learning tools will continue to be useful in simplifying the complexity
of electricity markets.
Using a simple prediction model, Davis and Hausman (2016) documents how the sudden closure of a
nuclear power plant near Los Angeles affected the supply of electricity. As a first-order effect, it implied the
reduction of a substantial amount of low-cost power, increasing costs of production substantially. Among
second-order effects, the closure created increased transmission constraints in the California electricity mar-
ket. The paper the proposes a methodology in the spirit of Mansur (2008). It proposes a flexible regression
approach to predict whether a thermal power plant will produce when aggregate thermal production is at
certain levels (using bins). This prediction model is based on pre-period data and it can be used to predict
what the first-order effect of the nuclear outage should be, under the assumption that the closure increased
aggregate thermal production approximately one-to-one. Comparing this to a model of ex-post behavior, the42Additionally, equation (23) becomes much more complex if firms do not take transmission bottlenecks as given.
80
authors find that power plants in the south of California disproportionately increased their production due to
congestion. The paper also confirms that two plants in the south exhibited the opposite behavior, constrain-
ing their output more than predicted. These power plants were subsequently fined for market manipulation.
4.3 Renewable power and the energy transition
Electricity markets are undergoing substantial transformations with the increased presence of renewable
power, mostly wind and solar power. Renewable power has been growing in part thanks to subsidies, credits,
and other incentive schemes such as explicit quantity goals in the form of Renewable Portfolio Standards.
As of today, their costs have plummeted and they are expected to emerge as a leading source of electricity.
The literature in Industrial Organization has helped understand several aspects of this transition. What
are the impacts of growing renewables on electricity market outcomes and the environment? What are the
costs of expanding renewables? How do dynamic costs interact with renewable production? What has been
the role of learning-by-doing during the transition?
4.3.1 Estimating the environmental impact of renewables
Several papers have estimated the environmental benefits from increased renewables, which tend to displace
polluting technologies such as coal or natural gas. Cullen (2013) quantifies the emissions offset by wind
production in the Texas electricity market using a simple econometric framework. Cullen regresses the
output of thermal generators in the Texas market on wind output, which to a large extent is exogenous
and exhibits substantial random variation after controlling for demand, temperature, and congestion. Once
the responses of generators are estimated, he quantifies the implications for emissions by using calibrated
emissions rates as a function of the technology.
Kaffine et al. (2013) and Novan (2015) complement this analysis by using direct measurements of power
plant emissions from the Continuous Emissions Monitoring System (CEMS) when quantifying the impact
of wind power on Texas’ pollution. The two papers follow a similar approach. Kaffine et al. (2013) regress
total aggregate hourly emissions on hourly wind output under the assumption that it is exogenous, whereas
Novan (2015) instruments with wind speed and wind installed capacity. Both papers confirms reductions
in major pollutants (CO2, NOX , SO2) although the effects are somewhat smaller than in Cullen (2013).
Novan (2015) examines the possibility of increased ramping and startup of power plants leads to increases
in individual emissions rates. Fell et al. (2021) further quantifies the benefits from improved transmission
in ERCOT and find that the marginal value of wind can be up to 30% higher when transmission constraints
are relaxed. In another related study, Dorsey-Palmateer (2019) finds that conditional on the total quantity
of thermal power generated, increases in wind output are associated with a shift away from coal generation
and towards natural gas. The intuition behind this result is that the intermittency of wind increases the value
of natural gas generators that, relative to coal, are easier to ramp up and down. This shift away from coal
further increases the GHG mitigation benefits of wind power. Callaway et al. (2018) quantifies the value of
renewables across a much wider range of locations in the United States and document substantial variation
across regions, rather than within regions. This variation highlights the benefits from improved high-voltage
81
transmission. Understanding the bottlenecks in renewable deployment and transmission expansion is an
important topic for future research.
4.3.2 Estimating the market impacts of renewables
Renewables have had substantial impact on the operation of electricity markets. There have been concerns
about the intermittent nature of renewables, i.e., the fact that their output is not directly controllable and
uncertain, in contrast with more traditional technologies that can be used on-demand. When their intermit-
tency is taken into account, the value of renewables both in terms of emissions reductions and profitability
might be hampered (Joskow, 2011; Borenstein, 2012; Joskow, 2019).
Gowrisankaran et al. (2015) develops a social planner model of electricity dispatch to quantify the cost
of intermittency from solar output using data from Arizona. It focuses on understanding the role of reliability
constraints in the presence of intermittent output. Electricity markets need to be in constant equilibrium and
therefore have certain redundancy built around them to avoid blackouts and other forms of demand curtail-
ment. The model has demand curtailment based on the theoretical work in Joskow and Tirole (2007), which
allows for reliability and rationing, and solves for the competitive level of investment. The authors adapt
the model to the empirical application of solar production and quantify the amount of investment necessary.
The model compares rules of thumb for reliability compared to optimal rules. Through the empirical model
of electricity markets operation, it quantifies that the costs of intermittency are overstated when the rules
around reliability are not updated to reflect the distinct characteristics of renewable power. Therefore, the
paper highlights the need for flexible design of electricity markets in the presence of renewable resources, a
topic that needs further attention.
Bushnell and Novan (2018) takes a more reduced-form approach to measure the impacts of renewables
on the operation of electricity markets. It uses a retrospective regression analysis to quantify the impact
of solar production in the California electricity market. The study highlights that average wholesale prices
have decreased thanks to the large increase in utility-scale solar generation, but that the average effect masks
heterogeneous impacts. The analysis documents that prices have been raising in the evening, consistent with
substantial operational restrictions due to ramping constraints when the sun sets.
Cullen (2015) models explicitly the impact of renewables on the dynamics of electricity markets using
a structural model of plant operation. Using detailed hourly plant-level data from ERCOT, he estimates an
equilibrium single-agent dynamic model in which thermal power plants (coal and gas) decide whether to
run their power plant or not. Power plants are allowed to have minimum production levels, startup costs,
and ramping constraints. Due to the presence of startup costs, the introduction of renewables does not
necessarily displace coal production, but it can disproportionately displace natural gas, which has smaller
startup costs but also smaller emissions rates. Using counterfactual analysis, he concludes that carbon taxes
need to be sufficiently large to displace the dirtiest generators, as shown in Figure 3. This exercise highlights
the importance of accounting for dynamics when analyzing the energy transition.
Cullen and Mansur (2017) shows that this kinked effect of carbon prices on abatement due to dynamic
effects is borne by the data. The emergence of cheap natural gas due to hydraulic fracking can be seen as
an analogous cost shock to carbon prices: it makes natural gas relatively cheaper than coal. Using a flexible
82
Figure 3: Dynamics in electricity markets can substantially affect policy counterfactuals
Note: A static model of electricity generation can be biased. It will overstate the production responses from
inflexible generators at low prices, and understate their responses. The dynamic response function is instead
kinked. Source: Cullen (2015).
regression approach, the paper shows that emissions from fossil fuel power plants have only substantially
declined once natural gas prices have made natural gas a more economical technology. Their estimated
semi-parametric emissions response to natural gas prices exhibits the same kink as Figure 3.
The presence of these dynamics affects in turn the market price and the profitability of power plants,
which impacts their long-run investment strategies. To understand the energy transition, it is crucial to get at
these longer run dynamics as well. However, one challenge with the modeling of start up costs in electricity
markets is that they make the notion of a competitive equilibrium difficult (O’Neill et al., 2005). Cullen
and Reynolds (2015) proposes to adapt a traditional Hopenhayn (1992) model to account for the fact that
the presence of startup costs and minimum production limits leads to non-convexities in the production set.
The key to their insight is that the model allows firms to have non-convexities, but firms are in themselves
assumed to be infinitesimal. This framing of the problem allows to solve for a competitive equilibrium,
including investment, while still preserving the main economic impacts of startup costs in profitability.
Linn and McCormack (2019) also develops a computational model to simulate the operational and in-
vestment decisions in the electricity market. It uses the model to understand the impact of environmental
regulation, renewables, and hydraulic fracturing in the exit of coal power plants. The proposed approach
relies on backward induction to solve for the dynamics. To account for startup costs, the dispatch model
puts constraints on the number of hours that coal power plants need to operate within a day. One advantage
of this approach is that, by simplifying the dynamics to a daily problem, the model can have more detail and
arbitrary states, something that is often restricted in dynamic models due to the curse of dimensionality.
83
Finally Qiu (2020) considers the possibility that ownership of wind assets can convey an informational
advantage in wholesale markets. In the U.S. Midwest ISO (MISO), firms owning wind assets historically
received higher-quality wind forecasts than did firms without wind assets. Accordingly, Qiu (2020) shows
that these firms’ bids into wholesale auctions are more responsive to realized wind generation than the
bids of firms who are relatively uninformed. The paper then develops a model of MISO auctions in the
spirit of Hortacsu and Puller (2008) to study a counterfactual in which all firms receive the high-quality
wind forecasts. Qiu (2020) finds that this information provision induces firms to bid more competitively on
average, yielding an average market price reduction of 3%.
In all of these studies, a key ingredient is the fact that renewable power is volatile and, therefore, leads to
an increase in the need to startup and shutdown power plants. However, the impact of volatility is expected to
be mitigated by the raising emergence of batteries. Optimally using batteries is a strategic dynamic problem
and therefore IO tools lend themselves well in this application. How to design and evaluate markets with a
large presence of renewables and batteries is an area in which more research is needed. Recent papers in IO
have started to explore this dynamic problem in markets in which there is a large presence of solar power,
which can lead to important price fluctuations and even negative prices. Batteries act as arbitrageurs to help
flatten valleys and peaks.
Karaduman (2020) analyzes the interaction of battery expansion and market power in the Australian
national electricity market (NEM). As an interesting observation, it highlights that, in markets with imperfect
competition, the incentives of battery operators will be distorted even if they act competitively, as market
prices do not send the right signals. He builds a Markov-Perfect Nash Equilibrium model in which each
period is a day (with several hours). The dynamic component of the game is to decide the battery level at
the end of the day based on the expectations going forward. In the counterfactuals, the paper documents the
discrepancy between private and social incentives, which is present when batteries do not internalize their
effect on market prices for other participants. This gap between social and private incentives is increased in
the presence of market power.
Butters et al. (2020) takes a complementary single-agent dynamic approach to the problem, studying the
impact of batteries in California, a market with substantial solar production. It solves for the competitive
equilibrium in battery storage, while taking the rest of the market as given with a competitive supply curve.
This simpler competitive structure allows them to consider endogenous entry of batteries, which is key to
simulate how the market will evolve under alternative subsidy policies. The counterfactuals focus on the
understanding the impact of alternative support policies, such as renewable portfolio standards and storage
adoption mandates.
4.3.3 Renewables and learning-by-doing
Another dynamic aspect in the adoption of renewable power that speaks to the core of the IO literature
is the presence of learning-by-doing in their development and installation. The costs of renewable power
have plummeted in recent years and are expected to continue to fall. Early work in the learning-by-doing
literature studies the drivers that are behind the cost improvements observed in solar (Nemet, 2006) and wind
(Nemet, 2012). These papers take a parametric approach to decomposing the drivers behind the observed
84
marginal cost trends. Anderson et al. (2019) explores to which extent learning is confined to the firm, or it
has spillovers to the industry, which can have different implications for optimal subsidy design.
Whether learning-by-doing happens within the firm or there are spillovers, public subsidies can speed
progress along the learning curve by incentivizing demand. Gerarden (2019) uses state-of-the art methods
in the dynamic games literature in IO to understand the role of demand-induced technological progress in
solar photo-voltaic (PV) power. The goal of his empirical model is to quantify the role of public subsidies
in stimulating technological improvements. In his model, firms’ actions are limited to investing to improve
their technological frontier. Subsidies increase the incentives to invest by shifting demand outwards. He
estimates the model using detailed firm-level data on the technological efficiency of solar panels, combined
with market level information on subsidies, sales (in Watts), and prices. The estimation follows the dynamic
games literature and uses conditional choice probability estimation coupled with forward simulation to con-
struct a pseudo maximum likelihood estimator. Using a moment-based Markov perfect equilibrium concept
(Ifrach and Weintraub, 2016), the paper then presents two counterfactuals: one in which subsidies only in-
centivize demand versus one in which increased demand also leads to increased incentives to invest. The
main findings of the paper is that subsidies have been essential at expanding demand for solar panels during
the study period (2010-2015), leading to substantial environmental benefits (in the order of $15 billion).
Additionally, demand-induced technological progress increases the environmental benefits by over 20%.
4.3.4 Renewables and demand-side dynamics
Other papers have focused on the dynamics on the demand side of renewable adoption. De Groote and
Verboven (2019) microfounds a dynamic demand system using a single-agent optimal stopping problem in
the spirit of Rust (1987). In the model, consumers decide when to optimally invest taking into account the
evolution of future costs and subsidies. Additionally, consumers may discount the future benefits of solar
panels too much, in line with the literature in fuel efficiency and vehicle adoption discussed in section 3.1.1.
This opens the possibility that investment-based subsidies, which occur at the moment of installation, are
preferred to production-based subsidies, which are materialized over a longer span of time, typically 20
years.
The study analyzes a program in Belgium between 2006 and 2012 using detailed level at the local market
to allow for consumer heterogeneity. The intuition behind the identification of the discount factor, which
is typically difficult to identify, lies on the fact that predictable changes in subsidies impact the expected
value of waiting but do not impact current utility (Magnac and Thesmar, 2002). Using predictable variation
in future solar subsidies, the discount factor is estimated to be of around 15 percent in their application.
The authors interpret this discount factor as large when compared to market rates, and quantify that solar
programs could be substantially cheaper for the government if they directly subsidized the upfront costs of
investment.
Langer and Lemoine (2018) examines residential (rooftop) solar PV subsidies in California by focus-
ing on deriving the optimal subsidy schedule. The paper expand the theoretical literature by considering
optimal subsidy decisions when consumers are forward-looking and costs are trending down, highlighting
an important trade-off in the setting of optimal subsidies. On the one hand, optimal subsidies could follow
85
a declining path, taking advantage of improved technological progress. On the other hand, optimal sub-
sidies could follow an increasing path, taking advantage of early adopters, who, via selection, have high
willingness to pay. Using detailed installation-level data from California Solar Initiative, coupled with elec-
tricity rates and demographics, the authors estimate the underlying valuation of consumers for solar panels.
Counter to observed programs in practice, the optimal subsidy is increasing.
Feger et al. (2017) also studies the adoption of solar panels using using a unique matched dataset on en-
ergy consumption, income, wealth, solar panel installations, and building characteristics for 165,000 house-
holds in Switzerland in 2008-2014. The focus of the study is on the distributional implications of solar early
adoption, which tends to be correlated with income (e.g., see also Borenstein (2017) for a distributional
analysis of the California Solar Initiative). The early adoption of solar panels by wealthy households can
lead to what is often called a “death spiral” in utility rates. By investing in solar panels, households are able
to reduce their utility bills and contribute less to paying the fixed costs of the electricity grid. The authors
examine to which extent a social planner can change the parameters of electricity rates to trade-off efficiency
and equity concerns, combining a dynamic adoption model with a regulator’s objective to maximize welfare
subject to equity concerns.43
Feger et al. (2017) uses the data to estimate structural models of energy demand and PV installation.
Households are forward-looking and solve a dynamic optimal stopping problem in their solar panel adoption
decision. The authors estimate the parameters of their energy demand function using a geographical bound-
ary regression discontinuity design that exploits price variation at spatial discontinuities between electricity
providers, to address the endogeneity of energy prices and fees. The combined electricity consumption
and PV adoption models allow them to simulate the effect of energy tariffs and subsidies on PV adoption,
welfare, and redistribution. In the context of Switzerland and in contrast with results from California (Boren-
stein, 2017), the paper finds that the impacts of solar adoption are not regressive on average, although they
are highly uneven. Whereas a subset of rich households benefit from solar, other rich households do not
adopt and pay high electricity costs due to their high levels of consumption.
4.4 Electricity demand
Electricity demand, particularly at the residential level, is known to be substantially inelastic, due to the
limited ability or incentive of households to respond to electricity prices. A growing literature is concerned
with the estimation of the elasticity of demand as well as experimenting on the best ways to get demand to
respond. Increasing the response from demand can become particularly valuable in the presence of growing
renewables and extreme weather events. Whereas technological advances in batteries are facilitating the
response to such emerging challenges, demand response can contribute to make the transition more efficient.43A similar paper is Wolak (2016), which studies the trade-off of setting water rates when considering equity and water balancing
goals. Wolak (2016) uses a structural approach to model the optimal non-linear tariff of a utility that needs to constrain water usein the presence of consumer heterogeneity and equity concerns.
86
4.4.1 Estimating the elasticity of electricity demand
A first step at understanding the scope for demand response is to credibly estimate the elasticity of electricity
demand. When thinking about the elasticity of demand, it is important to keep in mind the horizon: short-
, medium-, and long-run. The short-run elasticity of electricity demand tends to be very small, in part
because short-run retail prices tend to exhibit no variation for many residential consumers, thus preventing
any response to wholesale price shocks. The medium- and long-run elasticities of electricity demand can
be larger due to the ability of households to respond to high electricity prices via capital investments, habit
formation, or adoption of substitutes (e.g., heating or cooking modes). In all cases, the empirical difficulty
lies in finding credible and exogenous variation in prices that can enable the estimation of these effects. In
this subsection, we review papers that use observational data. In the next, we examine the experimental
literature.
Reiss and White (2005) uses a structural model and household-level survey data from the Residential
Energy Consumption Survey (RECS) in California to estimate the elasticity of electricity demand. The
key to the identification strategy resides in exploiting the presence of non-linear pricing. As is common
in many utility settings, consumers face higher marginal prices as their electricity consumption increases.
This increasing marginal price schedule implies that an OLS regression of consumption on marginal price
cannot consistently estimate households’ demand elasticity. By construction, consumers in the higher tiers
are those who consume more. Therefore, regressing consumption on prices without taking into account the
source of price variation would produce biased results.
Reiss and White (2005) exploits the different tiers as a source of price variation. Variation of behavior
within a tier helps identify the role of other covariates in shifting electricity demand. To control for the
endogeneity of the tier at which consumers are at, and the presence of bunching, it uses the selection term
from the structural model to control for the bias that would otherwise be present in the OLS regression.
Intuitively, consumers on the lower tiers have lower unobserved demand shocks that one needs to control for,
and these are reconstructed with structural assumptions. The main assumption is that these random shocks
(e.g., a month with higher than usual need for electricity) induce consumers to face different prices at the
margin. Reiss and White (2005) finds that the elasticity estimated using this approach is -0.39 on average.
This elasticity is relatively high for electricity consumption and should be interpreted as medium- to long-
run. Interestingly, the paper is also able to identify heterogeneity in the elasticity estimates. Intuitively, it
finds that households with either heating or air conditioning are more responsive to the level of prices, with
elasticities of -1.02 and -0.64 respectively. It also finds that lower income tends to be associated with more
responsive behavior (-0.49 vs. -0.29 for the lowest and highest income quartile).
A key underlying assumption of Reiss and White (2005) is that consumers are aware of the marginal
price they are facing when choosing their electricity consumption. Taken literally, the framework predicts
that consumers should be bunching around the quantities in which there is a change in the price schedule.
However, this bunching is rarely observed in practice. Ito (2014) examines to which extent the lack of
bunching in consumption could be due to the fact that consumers respond to average prices, as opposed
to marginal prices. To reduce concerns about the comparison between households in different areas, he
focuses on studying consumer behavior in Orange County, CA, which is served by two utilities (Southern
87
California Electric and San Diego Gas & Electric). He then focuses on the consumers surrounding the
border of the two regions. Interestingly, the border cuts through several cities, as opposed to coinciding with
other administrative borders.
Using detailed monthly electricity billing records, Ito (2014) shows that households at the two sides
of the border are comparable, but that they face substantially different non-linear tariffs during the study
period. He then uses a differenced panel regression approach with city-by-time fixed effects in which the
dependent variable is the log of electricity consumption and the independent variable is the log of the indi-
vidual price. Because the individual realized price is endogenous to consumption in itself, he instruments
the observed price with the predicted price based on that households’ lagged consumption. In the context
of this regression, he finds that average prices are more negatively correlated with consumer consumption.
When both average price and marginal price are included in the regression, the average price seems to be
capturing all the explanatory power, thus lending support to the theory that consumers are not fully respond-
ing to marginal prices. Based on this estimation, he finds that the elasticity of households to average prices
is in the order of -0.08.
Shaffer (2020) uses a similar design using monthly billing data from British Columbia, exploiting a
tariff change in which one utility transitioned into a two-tiered tariff but the neighboring utility did not.
Different than Ito (2014), Shaffer observes bunching in the distribution of electricity consumption. Using a
variety of bunching estimators, he is able to estimate an elasticity of around -0.04. When considering the
differenced regression framework as in Ito (2014), the marginal price, as opposed to the average price, is the
one with the most explanatory power. How to reconcile these results? Shaffer develops a structural model
in which there are three types of consumers: those who respond to marginal prices, those who respond to
average prices, and those who believe that the marginal price applies to all consumption. He uses an indirect
inference approach to find the share of types by consumption decile that are most consistent with the reduced
form regression evidence. He shows that the effects appear to be driven by a small share of consumers who
behave as if the marginal price affected all consumption, and thus have very high incentives to bunch.
The previous papers are eminently based on exploiting differences in pricing due to the tariff structure,
either as a cross-section or with some panel variation. Can one estimate the elasticity of electricity demand
in the absence of non-linear pricing? Deryugina et al. (2020) exploits a time series change with the shift
towards municipal choice aggregators in the state of Illinois, which led to sudden and significant reductions
in the prices of electricity. A key challenge in the identification of the price effects is that communities
that select into community aggregation plans are not exogenous. To alleviate these concerns, they match
communities to similar comparison controls based on detailed electricity usage data. Using this flexible
difference-in-difference matching approach, they estimate that the price elasticity of demand grows from -
0.09 in the first six months to -0.27 two years later. This highlights that consumers can respond to permanent
decreases in prices in the medium-run.
The previous papers focus on estimating the medium- to long-run elasticity of residential consumers.
Absent explicit pricing programs, most consumers do not face variable prices in the short-run. Fabra et
al. (2021) takes advantage of the fact that Spain defaulted a large share of residential consumers into real-
time pricing. Using detailed smart meter data together with wholesale electricity prices, the paper estimates
88
individual elasticities using a simple log-log regression, instrumenting the price of electricity with wind
production forecasts. The results show that the short-run elasticity of consumers is essentially zero in their
setting, in which consumers are often unaware of the price, and price variation in wholesale prices is rela-
tively modest when compared to the final electricity price that consumers face.
Finally, there is a smaller literature estimating the elasticity of commercial and industrial consumers to
electricity prices by exploiting the fact that this type of consumers often face variable prices in the form
of mandatory time-of-use pricing (Jessoe and Rapson, 2015) or critical peak pricing (Blonz, 2016). Both
papers estimate that commercial and industrial consumers are responsive to these programs. In terms of
quantifying their elasticity of demand, Blonz (2016) finds, however, just a modest short-run elasticity to
peak pricing in the order of -0.11.
4.4.2 Experimental evidence on demand response
A growing literature in demand response exploits the presence of smart meters and the possibility of ex-
perimenting with pricing designs that more actively engage consumers. It would be impossible to cover all
papers in detail, thus in this section we present a small selection of papers.44
Jessoe and Rapson (2014) estimates the demand responses of households to a pricing experiment by part-
nering with a utility in Connecticut. The pricing experiment features critical peak pricing, i.e., in days with
difficult operational conditions, households face much larger prices. In the treatment arms, these scarcity
conditions are communicated to households either the day before or 30 minutes in advance. Additionally, a
subset of treated households receive an informational display. The main finding is that consumers substan-
tially respond to these pricing events when they have the informational display, and particularly so if they
are told a day in advance, with an elasticity of almost -0.20 for the most responsive group. Is this elasticity
a truly short-run response? Or does it reflect potential medium-run adaptation? Jessoe and Rapson (2014)
finds evidence that consumers responded in the medium-run, by changing their behavior more generally.
Increasing the short-run elasticity of demand can be critical to reduce operational costs and to miti-
gate market power. Do these pricing events meaningfully engage real-time responses? Or is automatic
technology necessary? Bollinger and Hartmann (2020) highlights that peak-pricing experiments have one
limitation: the day of the pricing event is not random. Instead, it occurs during periods of high electricity
demand. Thus, the estimated effect can be a combination of both short-run and medium-run responses.
The paper proposes to use a control function approach to account for the endogeneity of pricing events and
better capture the non-linear patterns of demand consumption shocks and their correlation with price. Using
experimental data, the findings show that in the absence of technology, the short-run elasticity appears to be
close to zero even for households with informational displays. Technology appears to substantially increase
the elasticity of consumers. How to enable and evaluate the roll-out of technology-assisted demand response
is likely to require more research attention in the near future.
A challenge of many dynamic pricing experiments is that the participation rates are rather modest, thus
questioning the ability to scale up such programs. What would happen if consumers were defaulted into such
tariffs instead? Fowlie et al. (2020) studies a large randomized controlled trial in the Sacramento Municipal44The interested reader can find a recent review of experimental papers in electricity demand in Harding and Sexton (2017).
89
Utility District (SMUD) between 2011 and 2013. One of the major contributions of the paper is to identify
the effect of defaults in the adoption of dynamic pricing by explicitly randomizing the default tariff in
the two treatment arms, one with a default flat tariff and one with a default critical peak pricing scheme.
In the experiment, over 90% of households decided to keep the dynamic tariff, while only 20% would
have actively opted in. Even if the average response of those who were defaulted is lower on average, the
aggregate effect can be substantially larger once participation is taken into account. The results suggest that
defaulting consumers into explicit dynamic pricing tariffs, when politically feasible, might be a good avenue
to increase the scope for demand response. The experiment in itself allows them to quantify the elasticity to
critical peak pricing events, which they estimate to be -0.075, in line with the previous literature.
Other papers in the literature have explored the potential of using non-price signals to engage demand
response in moments of critical need. Ito et al. (2018) compare critical peak pricing events to messaging en-
couraging consumers to reduce their electricity demand in the aftermath of the Fukushima accident in Japan.
They find that the response to pricing incentives is more pronounced, but that informational campaigns in
moment of scarcity can also lead to demand responses. Andersen et al. (2019) also finds evidence consistent
with this pattern in a randomized controlled experiment in Denmark. It finds that customers respond to
both encouragements to reduce or increase power, by shifting their consumption, and that the patterns are
more pronounced in the presence of monetary incentives but still there in the presence of an environmental
motivation.
Whereas some of these experiments are encouraging, with the exception of Fowlie et al. (2020), most
of them are focused on only a small subset of households who select into the experiments. Given the need
to increasingly engage consumers to respond to extreme weather events and extreme operational conditions
as well as the growing ability to respond to prices algorithmically, further work in this area will be needed
to understand how to scale-up these pricing mechanisms. These responses in demand will be critical to
encourage to ease the adaptation to increasingly stressful conditions in the grid due to climate change, e.g.,
as experienced in California in 2019 and 2020 during the wildfire seasons or due to extreme cold in Texas in
2021. Understanding how demand can be flexibly deployed while ensuring safe conditions and mitigating
increased inequalities will be of utmost importance.
4.4.3 Competition in retail electricity markets
Given the low demand elasticity by residential customers, a natural IO question in the context of residential
demand is the extent to which firms compete to attract customers. First, it is important to note that compe-
tition in the retail side of electricity has lagged behind the liberalization of generation, both in the United
States and Europe. This is already an indication that generating competition has been difficult. Second,
even conditional on introducing liberalization of retail tariffs, growing competition among firms was modest
in its first attempts, due to limited attention and also the difficulty of independent retailers to hedge their
contracts, which often leads to attrition in fringe retail competitors.
Because electricity retailing has often been in the hands on the incumbent distribution company, a big
hurdle to increase competition is the presence of consumer inertia and switching costs. Hortacsu et al. (2017)
explores the presence of these barriers in the context of the Texas electricity market, ERCOT. The retail
90
market for electricity in ERCOT was liberalized in 2002. Consumers were defaulted into the incumbent
company, but they had the option to visit a website to change providers. Even though prices of the incumbent
tariff were higher and that savings could be substantial (around 8% per year on average), consumers took
a long time to switch. Four years after the policy change, over 60% of consumers were still on the default
tariff.
The key contribution of the paper is to separately identify how much of the persistence is driven by
search frictions and inattention vs. a preference for the incumbent provider. This separation is important, as
it might affect the shape of public policies intended to make the retail market more competitive. They iden-
tify these two sources via a structural model of consumer choice with two stages. First, consumers decide
whether to consider switching retailers. Second, conditional on going to the website, they can chose among
all retailers. The identification strategy combines parametric assumptions on the probability of considering
alternative choices and the utility provided by each retailer with unique flow data on the amount of con-
sumers transitioning between retailers. Intuitively, the parameters that inform the probabilities of switching
and the households’ utilities need to be consistent with the observed flows. The authors estimate the model
using a GMM approach in which they match observed flows to predicted flows as a function of parameters.
To allow for persistence, the model allows consumers to be more attached to the incumbent the longer they
stay with the default provider. The study highlights that small interventions that increase switching and/or
reduce the preference for the incumbent would substantially increase consumer welfare.
The previous paper documents the stickiness in consumer behavior in electricity retail markets, but
remains silent about the competition implications of such frictions. Duso and Szucs (2017) quantifies the
degree of competition in the German retail electricity market by examining the pass-through of wholesale
costs to retail tariffs during the period of 2007-2014. It exploits panel variation in Germany’s competitive
structure and wholesale final prices to examine how cost shocks are passed through to consumers. Using a
fixed-effects regression model, the findings show evidence that competitive retailers exhibit larger average
pass-through than incumbents (0.70 vs. 0.50), consistent with the presence of switching costs. The results
also show evidence consistent with the market becoming more competitive over time, at least as explained
by the presence of a pass-through that gets closer to unity in all segments.
Byrne et al. (2019) performs a field experiment to better understand the competition impacts of search
frictions in the context of the electricity market of Victoria, Australia. In this market, low-income households
are able to receive subsidized electricity rates, but consumer reports suggest that the pass-through of the
subsidy is incomplete, leading to concerns about competition. The authors perform a randomized audit
study in which actor-consumers pretend to be searching for better electricity rates. They show how these
experiments can help identify the degree of discrimination and competition in the market in a setting with
negotiated prices and search costs.
Further work is needed to understand how recent changes in the supply side of electricity market are
affecting competition in the retail market. It is also critical to understand how the increased used of dynamic
pricing schemes will impact competition and consumer surplus. Due to the likely increase in price volatility
and costs, the distributional implications from retail competition can also be a substantial, another important
area of future work.
91
4.5 Cap-and-trade regulation in electricity markets
The electricity sector has traditionally been one of the largest emitters of airborne pollutants, either in the
form of local pollutants (particulate matter, SO2, and NOX ) or global pollutants (CO2 and methane), to-
gether with other toxic agents (mercury). Thus, power plants have been subject to numerous environmental
regulations, initially due to local pollution concerns and more recently due to climate change concerns. Sev-
eral tools have been used for the regulation of harmful emissions. Command-and-control approaches and
state-level standards have been a common tool, but cap-and-trade markets have gained prominence in the
regulation of emissions since the start of the SO2 trading program in 1995.
Electricity markets provide a unique opportunity to study cap-and-trade regulation because there is a lot
of institutional knowledge and data about both the product market (production of electricity) and emissions
market. This is not necessarily true in other sectors, in which emissions regulation is well observed but
the modeling of the supply side can be challenging due to the presence of varying products and technolo-
gies, unobserved heterogeneity, and limited frequency of data. Therefore, one can consider the strategic
interactions between firms and markets with a great amount of detail.
There is a vast literature examining the design and performance of these markets. See, for example, a
retrospective of the SO2 trading program in Schmalensee and Stavins (2013), which regulated emissions
of SO2 from power plants in 21 eastern and Midwestern states.45 There is also a large body of literature
studying the properties of cap-and-trade markets. We provide here a necessarily incomplete overview of
empirical papers given the breath of IO topics that can be explored using the data from these markets.
4.5.1 Emissions markets in the US electricity sector
The maximization problem faced by electricity generation firms in the presence of cap-and-trade regulation
can be summarized as follows:46
maxqit,ait
ptqit − Ci(qit, ait)− τt × (ei(ait, qit)− φit(qit)) , (28)
where i indexes the firm, t a time period (e.g., an hour), pt is the wholesale electricity price,C are production
costs, which can depend on quantity as well as abatement effort, τt is the permit price, ei is the emissions
function, which can be reduced with abatement effort, and φit is a free allocation of permits that reduces
the environmental costs of firm i. The free allocation can be a fixed amount of permits, what is often
called “grandfathering,” or potentially an endogenous function of the firm’s output qit, what is often called
“output-based updating.”
There are many IO questions that have been studied in this setting, either theoretically or empirically.
Are these markets functioning as expected? Are there market power concerns in both, or inefficiencies
spilling from one market to the other? What is the role of the free allocation φ? What about in a dynamic
sense? What are the interactions between the product and the emissions market?45See also Schmalensee et al. (1998) for an early economic review of how the market performed in its first years of operation.46For notational simplicity, the equation here represents a firm with a single power plant, in practice firms maximize profits
summing over several power plants.
92
Carlson et al. (2000) presents a first assessment of the abatement efficiency of the SO2 market by com-
paring observed outcomes to an ideal least-cost solution. The study highlights that many of the seemingly
large gains from trade from the introduction of the SO2 market were thanks to the advances in the ability to
burn low-sulfur coal at existing generators.47 It then focuses on computing the efficiency of the market tak-
ing holding the low-sulfur technology improvements constant. The paper finds that allowance prices were
close to marginal abatement costs, suggesting that the market was sending efficient signals at the margin.
However, the gains from trade are estimated to be smaller than the overall potential savings,finding that
the performance was in the order of 43% of an idealized least-cost solution. This suggests that there were
frictions in the market limiting the amount of trade.
Ellerman and Montero (2007) explores to which extent the SO2 market functioned efficiently over time
and comes with a more positive assessment. Using aggregate data, the paper documents the evolution
of Phase I of the SO2 market, highlighting that firms banked their 30% of the allowances for the more
stringent Phase II of the program. It then tests the temporal efficiency of these markets as in the exhaustible
resource literature covered in section 2.1.1. The paper calibrates a model of optimal banking and shows
that, under reasonable parameters, the observed aggregate trading patterns are not inconsistent with optimal
banking. The two findings are not inconsistent, as allowance prices could reach an intertemporal equilibrium
consistent with efficient trading but there could still be room in the market for unrealized trades in the
presence of frictions.
The presence of transaction costs interacts with a common feature in cap-and-trade markets: the free
allowances granted to the participating firms to cover at least part of their expected production, captured
by φit. The permit allocation can lead to distortions in the market if there are transaction costs or if the
allocation is a function of endogenous factors; e.g., if it is tied to production.48 In equation (28), the alloca-
tion φ will clearly distort decisions if it is a function of output (statically or dynamically in a more general
framework). It can also lead to distortions if trading permits is costly and thus the effective emissions costs
depend on trading positions and not just the permit price τt (Stavins, 1995).
Because the initial allocation is by definition almost always endogenous and correlated with output,
regressing output on the initial allocation will tend to suffer from reverse causality and lead to positive
estimates. Fowlie and Perloff (2013) exploits random allocations due to different allowance allocation timing
in the RECLAIM NOX market in Southern California and finds no evidence that the initial allocation, which
was independent of future behavior, affected firms’ choices.49
In a dynamic setting, a counterfactual approach has been used to explore the impacts of alternative
allocation rules. Fowlie et al. (2016), which is covered in more detail in section 5.1.2, examines the role
of the allocation in the cement industry. In the electricity market context, Dardati (2016) builds a dynamic47Even with the exercise of market power by railroads discussed above that put upward pressure on low-sulfur coal prices (Busse
and Keohane, 2008), abatement costs from the program were dramatically reduced thanks to these technological advances.48See Hahn and Stavins (2011) for a review of the distortions that can arise from the initial allocation and a review of the literature
and findings in several cap-and-trade markets.49In unpublished work, Reguant and Ellerman (2008) explores a non-linearity in the allocation of permits to coal power plants in
Spain to address the endogenity issue, also finding that the allocation did not affect firms’ behavior. However, the empirical variationis quite limited. Using a logit model of operation decisions, the paper also shows that electricity firms treated environmental costsin the same way as revenues, lending further support to the lack of distortions.
93
model in the style of Hopenhayn (1992) to show how the allocation, which in some markets is conditional
on staying in the market, can affect dynamic entry and exit decisions. The paper builds an empirical model
of the SO2 market and shows that giving permits to new entrants, as opposed to letting incumbents keep
them, can substantially impact productivity by encouraging more entry.
Toyama (2019) explores the extent to which market frictions, such as transaction costs, could have
limited trading using a dynamic structural model. The model abstracts away from strategic market power,
given the previous evidence, and models firms in a single-agent dynamic competitive equilibrium. In the
model, firms are choosing both their abatement in the static problem in equation (28) as well as their decision
to bank permits for the next trading period. In the presence of transaction costs, banking can be a substitute
for trading. The paper estimates the heterogeneous effect of transaction costs, which depends on trading
position: those who buys (sells) permits tend to under- (over-) bank. The paper quantifies the welfare and
efficiency implications of this dynamic friction.
Chen (2021) also uses a dynamic single-agent model to explore departures from efficiency in the SO2
market. In the model, firms have biased beliefs about the permit prices, which can lead to inefficient dy-
namic decisions about storing allowances and can be a complementary source of trading frictions. Separately
identifying belief biases from other unobservables can be challenging in the absence of a metric to define
accurate beliefs. Methodologically, the paper shows how to identify the bias in the beliefs by taking advan-
tage of the fact that decisions of firms, other than their banking of permits, also affect compliance costs. In
particular, firms can use lower sulfur coal, and the value of using a permit today vs. reducing sulfur con-
tent should equalize in equilibrium. Failure of treating these two choices analogously can provide evidence
of lack of optimization. Empirically, the results suggest that firms under-reacted to price fluctuations, and
therefore it is consistent with firms’ expectations being biased towards lower price volatility.
Other papers have examined strategic distortions arising from the interaction between the permit and
the electricity market. Permit prices are an important cost shifter to producers in the electricity market, and
therefore they can have substantial impacts on product prices. Kolstad and Wolak (2008) explores these
strategic interactions in the South Coast Air Quality Management District (SCAQMD), which regulated
NOX emissions in Los Angeles. It finds that some power plants claimed to face higher than average prices
in order to circumvent regulatory scrutiny in the power market.
In recent years, cap-and-trade markets have also expanded to cover emissions related to climate change,
for example with the emergence of the EU Emissions Trading Scheme in Europe or the AB-32 cap-and-trade
market in California. Given the consolidation of these existing cap-and-trade markets for greenhouse gas
emissions, and the likely expansion of these markets to other territories, we expect the work in this area to
continue to grow.
4.5.2 Interactions between cap-and-trade regulation and regulatory regime
As explained in section 4.1, the United States is unique in its heterogeneous regulatory treatment of electric-
ity generation. Several papers have studied how this regulatory heterogeneity interacts with the functioning
of environmental markets.
94
Fowlie (2010) studies heterogeneity in compliance choice in the context of the NOX Budget Trading
Program, which set a cap of NOX emissions across several states in the Eastern interconnection. The paper
begins by documenting the technological frontier that establishes a trade-off between various capital in-
vestments, with NOX control technologies of varying effectiveness, and the purchase of emissions permits.
Using a discrete choice logit model, the paper then shows that generators in areas governed by traditional
cost-of-service regulation were likely to comply with the NOX regulations by investing in high-capital op-
tions that aggressively reduced pollution. In contrast, plants in restructured electricity markets were more
likely to comply by using less effective but cheaper technologies or by just purchasing permits in the NOX
market.50 Because the damages from NOx are not evenly distributed in space, the paper documents that the
disparity in capital investment choices between states with traditional utility regulation versus restructured
markets led to emissions reductions being relatively less ambitious in the high damage areas.51
Abito (2019) develops a structural model of the regulatory distortions arising from natural monopoly
rate-of-return regulation. Compared to Fowlie (2010) and Cicala (2015), the paper studies these interaction
in the context of the Phase I of the SO2 market, in which capital investments were more limited. The
paper focuses instead on the distortions in effort to reduce costs (Laffont and Tirole, 1993), showing that
firms are more inefficient in periods in which rates are being set. The distortions in effort due to asymmetric
information result in marginal abatement costs that are larger than in a first-best setting, therefore interacting
with the cap-and-trade market. A cap-and-trade market designed to achieve the first-best will fail to do so in
the presence of these additional distortions. The paper exploits the timing of rate cases to identify the cost
and effort function and infer the marginal abatement cost curve under optimal effort vs. no effort. It then
derives the optimal contract to achieve the second-best and compares it to a limited contract that allows firms
to either select into a fixed price contract or choose to be reimbursed for all of their costs, finding that this
more intuitive contract can deliver a substantial amount (65%) of the welfare gains achieved by the optimal
contract.
5 Environmental regulation of energy-intensive industries and natural re-sources
Energy markets and resource-intensive activities are heavily regulated due to the large presence of environ-
mental externalities. We have already covered many of the regulations aimed at the energy sector that have
the purpose of mitigating pollution. In this section, we explore applications that are not confined to energy
markets.
A large literature in environmental economics documents the health impacts of environmental external-
ities, how regulation has lead to a reduction in air pollution, and the economic value of such reductions.52
50This pattern is also documented in the context of sulfur dioxide abatement in Cicala (2015), which highlights how the distortionfits a model in which utilities have an incentive to over-invest in capital (Averch and Johnson, 1962).
51Regulated and unregulated areas affect each other through the equilibrium cap-and-trade price. Fowlie and Muller (2019) fur-ther documents the environmental consequences of the disparities in marginal abatement cost curves for regulated and unregulatedstates. The paper studies the benefits of moving towards alternative cap-and-trade designs that account for the geographical natureof marginal damages using a highly detailed geographical model of emissions damages based on Muller and Mendelsohn (2009).
52A leading example of such regulations is the Clean Air Act; see Currie and Walker (2019) and Aldy et al. (2021) for a more
95
Here we will focus the discussion on papers that use IO tools to study the impacts of regulation on manu-
facturing firms and their production choices. Due to the importance of trade in the discussions of how to
regulate environmental externalities in the manufacturing sector, a large related literature in trade has also
examined these questions (Copeland and Taylor, 2004; Cherniwchan et al., 2017).
5.1 Environmental regulation in manufacturing and resource-intensive sectors
The regulation of manufacturing industries is often uneven across countries. Researchers and policymakers
have long been interested in understanding the impacts of environmental regulation in the outsourcing of
polluting activities (Ederington et al., 2005; Levinson and Taylor, 2008). Such displacement of polluting
manufacturing sectors is often rationalized via an environmental Kuznets curve, in which lower income
countries are willing to take the more polluting activities.
Recent work in this area has focused on exploiting energy price variation to understand the potential
responses to stronger climate change regulation; e.g., in the form of carbon taxes. Ganapati et al. (2020)
uses confidential Census data between 1972-1997 to show the response of manufacturing output and prices
to energy costs. The paper’s goal is to understand whether consumers or producers bear the costs of such
shocks. It examines seven industries that reported quantities to the Census: boxes, bread, cement, con-
crete, gasoline, refining, and plywood. It first estimates the marginal costs of these industries by estimating
markups (Hall, 1986; De Loecker and Warzynski, 2012). It then performs regression analysis by looking at
the pass-through of marginal costs to prices. Because part of the marginal costs are due to energy inputs,
the paper uses a shift-share instrument based on the generation mix of electricity of different regions of the
US. The paper finds that producers do not fully shift the costs to consumers, which could either be a sign
of market power or a sign that firms are exposed to international competition and facing asymmetric cost
shocks.
Muehlegger and Sweeney (2017) examines pass-through in one of the sectors of study of Ganapati et
al. (2020), refineries. It exploits the fact that the fracking boom created differentiated shocks depending on
whether refineries were close to the new areas of exploitation. It finds evidence that the degree of pass-
through is largely affected by whether cost shocks are regional or common to all firms. This heterogeneity
in cost shocks is at the heart of the discussions surrounding leakage concerns.
5.1.1 Climate regulation and leakage
In the presence of global pollutants, such as greenhouse gases contributing to climate change, the outsourc-
ing of polluting activities not only has the problem of shifting economic activity and pollution overseas, but
it additionally has the challenge that pollution overseas is equally costly to those facing the increased regu-
lation costs. Therefore, the shifting of emissions outside of a jurisdiction does not imply reduced damages
to the regulating area. This problem can make unilateral environmental regulation quite ineffective.
“Leakage,” the term used for emissions going to unregulated jurisdictions, can be a serious barrier when
enacting climate legislation. International trade organizations and countries are recognizing more and more
comprehensive retrospective of the literature and its findings than what we can offer here.
96
the need for carbon adjustments that would tax emissions at the border, but the details on a transparent and
wide implementation are still not in place. For example, these regulations can be threatened by the difficulty
of verifying emissions from manufacturers in other jurisdictions, leading to potentially inaccurate border
adjustments that are difficult to measure. For this reason, the most common regulation to address leakage is
via output-based subsidies. Even in such cases, the measurement challenges to set individualized efficient
tariffs can be challenging (Fowlie and Reguant, 2018; Lyubich et al., 2018; Fowlie and Reguant, 2021).
Fowlie and Reguant (2021) seeks to inform optimal climate policy that takes into account leakage. It
derives simple formulas for optimal output-based emissions permit allocations under leakage, which depend
on the elasticity of domestic production, imports, and export, to carbon taxes. To estimate these elasticities,
it uses regression analysis to measure the impact of energy costs on manufacturing output using aggregate
publicly available data. The cost variation used in the analysis is related to Ganapati et al. (2020) and
Muehlegger and Sweeney (2017), exploiting the fact that different industries located in different areas face
different cost shocks with the fracking boom. Crucially, the fracking boom also generates differences be-
tween domestic and foreign energy costs, which are needed to identify the elasticities of output, imports, and
exports needed to inform the optimal output-based subsidies. This paper is more ambitious in its goal, as it
allows for a quantification of optimal subsidy policies in cap-and-trade regulation. However, the empirical
findings are limited by the use of aggregate data.
Current discussions on how to properly tax emissions at the border involve tracking of the inputs and
outputs that go into the production process that are generally abstracted away in IO modeling. To inform
these questions, our tools will need to double down on the “IO” and bring back more comprehensive input-
output analysis to the table.
5.1.2 Leakage in dynamic models
The regulation of emissions can also have dynamic consequences by affecting firms’ entry and exit decisions
and location choices. Mechanisms that are equivalent in a static model, such a grandfathering regime or a
carbon tax (Coase, 1960), can have substantially different dynamic consequences.
Fowlie et al. (2016) examines the dynamic leakage implications of alternative climate policies in the
Portland cement industry, which is responsible for 5% of GHGs emissions worldwide. The paper extends
the dynamic game model in Ryan (2012) to incorporate different mechanisms to tax firms for their carbon
emissions.53 The paper is focused on the comparison between a carbon tax, a carbon tax with grandfathered
permits, an output-based adjustment mechanism, and a border tax adjustment, in line with the policy tools
that are being considered to mitigate leakage in several markets, such as the EU-ETS or AB-32 in California.
Cement has the distinctive feature of being more easily transported by ship than by road. Therefore,
there is variation in the degree of international exposure to leakage. The paper estimates cement demand and
supply in several U.S. markets and performs counterfactual simulations of coastal vs. landlocked markets.
In coastal areas, even though they are more competitive, the social surplus-maximizing carbon price is
substantially below the standard Pigouvian tax due to the presence of leakage under output-based regulation.53Given the use of dynamic estimation techniques in both Ryan (2012) and Fowlie et al. (2016), the technical aspects of these
papers are already covered in more detail in the chapter “Dynamic Games in Empirical Industrial Organization.”
97
The most cost-effective policy for these markets is a border tax adjustment. For inland markets that are
not exposed to leakage, surplus-maximizing carbon prices can still be below the standard Pigouvian tax
due to the high degree of concentration. The paper also shows that for these concentrated, non-trade-
exposed markets, initial levels of taxes can transfer monopoly rents from producers to consumers while
having limited impacts on production.
Hsiao (2020) examines the scope for leakage in the palm oil industry, another major contributor to
climate change. Palm oil production drives widespread deforestation in Indonesia and Malaysia, including
in carbon-dense peatland regions. The paper explores the general equilibrium effects of border carbon
adjustments on palm oil imports when not all importing countries participate. Under leakage, lower demand
from countries that establish a carbon tax leads to a reduction in the market price of palm oil, which can
incentivize demand in other areas. The paper assesses the importance of international coordination and
commitment in addressing this leakage.
To get at general equilibrium demand effects, the paper builds a model of world demand for palm oil
with a country-level almost-ideal demand system for vegetable oils. On the supply side, it builds a dynamic
equilibrium model in which farmers make individual decisions about whether to deforest and, conditional on
deforestation, how much to plant. The paper uses Euler techniques to estimate the supply-side parameters
(Arcidiacono and Miller, 2011; Kalouptsidi et al., 2021). In the counterfactuals, the paper shows that both
commitment and coordination are crucial. Only when combined are these tools moderately successful at
reducing deforestation.
Recent decades have seen little success in curbing the speed and extent of deforestation, and further
work is needed on this important topic. IO tools are particularly well-suited for studying how deforestation
and land use choices respond to environmental regulation. While this handbook is necessarily incomplete,
papers like Scott (2013), Souza-Rodrigues (2019), Assuncao et al. (2021), and Cole et al. (2021) contribute
to our understanding of land use and deforestation and are great examples of how to apply IO tools to address
these pressing questions.
5.2 Imperfect monitoring and the enforcement of regulations
A growing literature examines how firms respond to environmental regulation and monitoring using IO
tools. Properly enforcing environmental rules is critical to ensure that the expected gains and losses from a
given policy are realized. Whereas many of the previous papers assume that compliance with regulations in
the jurisdiction of study can be ensured, compliance is not always the norm in practice. In section 3.3, we
discussed examples of firms violating fuel economy standards by mis-reporting their vehicles’ performance.
Here, we highlight other recent work examining non-compliance with environmental regulations.
5.2.1 Evidence of cheating in environmental settings
Properly understanding firms’ strategic responses to enforcement and monitoring of environmental regu-
lations can be difficult due endogeneity concerns, such as higher inspections in days with high levels of
pollution. To address this challenge, Zou (2021) exploits quasi-experimental variation in the auditing sched-
ule of emissions monitors to show reduced-form evidence that manufacturing plants appear to endogenously
98
respond by shifting pollution away from the days in which they expect to be monitored. It shows that, for
monitoring stations in which the day of auditing is quasi-random (determined on a six-day schedule), satel-
lite measures in the surroundings of the monitors appear to record lower pollution levels during the days in
which monitoring occurs. The paper then addresses the question of what the mechanisms are that could lead
to lower pollution in days in which emissions are officially recorded. It shows that regulators themselves
are 10 percent more likely to issue an advisory on monitoring days, which implies that local authorities
themselves are being strategic to avoid falling out of compliance.
Oliva (2015) examines cheating at car smog checks in Mexico and finds evidence of bribing at the test
centers. The test exploits the institutional details of cheating, which rely on using the measurements from
the previous car in the testing lane. Serial correlation with testing outcomes from the previous car can be
used as a reduced-form test for cheating. In addition, the paper constructs a maximum likelihood estimator
based on the transition probabilities of passing the test. Exploiting the fact that there is substantial variance
in the outcomes of the test, even conditional on the same car, the identification assumes that drivers do not
learn about their car’s pollution after failing a test. At that point, they are faced with the decision of paying
a bribe to pass, or re-taking the test, which they can do only once for free. The fact that the second test is
free, but not the third one, is helpful to identify the parameters in a two-period model in which drivers can
decide to test the car, with or without bribing, and re-test if they fail. The estimation is based on matching
the observed probability of passing the test at different observable stages of the process (passing/no passing,
retesting, etc.). The estimation finds that 10% of cars pass the test via cheating, an estimate that is confirmed
using a simpler approach that compares centers with and without evidence of cheating in their tests. The
paper then compares situations with and without bribing to quantify the welfare implications of cheating.
5.2.2 Structural models of enforcement
The previous two papers show evidence of cheating but do not model explicitly the regulatory costs of
designing, monitoring, and enforcing regulations. Recent work has explored that interaction structurally.
Kang and Silveira (2018) builds a structural model to understand the regulatory game between the regu-
lator and the manufacturers when it comes to environmental enforcement. The paper focuses on the trade-off
between transparency and discretion faced by regulators when setting their rules. Whereas discretion might
allow regulators to fill in gaps in otherwise incomplete enforcement policies, it has the risk of leading to
regulatory capture and disparate outcomes at otherwise similar sites. The paper’s application is enforcement
of the Clean Water Act in California. It first documents that penalties for violations of standards appear
to exhibit discretion. The paper then builds a model of optimal environmental regulation in which firms
are heterogeneous in their costs of abatement but their type is unknown to the regulator. It then sets up an
optimal regulation problem a la Laffont and Tirole (1993) that it brings to the data. To estimate the regula-
tor’s cost of enforcement, it exploits a regulatory change in 2006 that made monitoring easier and increased
the regulator’s resources. The empirical findings suggest that discretion provides significant value in this
setting.
Blundell et al. (2020) examines the dynamic implications of monitoring and enforcement in the context
of air pollution regulation by the U.S. EPA. It starts by noting that the regulation currently in place has
99
a dynamic component, with repeat offenders being subject to larger fines and higher scrutiny. Thus, the
mechanism has an escalation component, by which offenders placed under the high priority violator cate-
gory end up paying much larger fines if they are again out of compliance in a future period. The authors
build a dynamic model in which plants need to chose the optimal level of abatement investment given the
predetermined schedule of fines set out by the regulator and the presence of dynamic enforcement. The
dynamic model is used to estimate the costs of investment and the disutility of paying fines and being placed
under high violator status. To allow for flexible heterogeneity, the authors use the estimator in Fox et al.
(2011) that allows for discrete types. They then consider counterfactuals in which the regulator does not use
dynamic penalties. Their estimates imply that the dynamic structure is effective at preventing offenses while
reducing the need for collecting fines. A static policy would require much larger fines to ensure the same
level of enforcement.
The enforcement of environmental regulations can potentially be more critical in developing countries,
in which the baseline levels of pollution tend to be larger and the ability to monitor could be harder. Using
a dynamic model combined with experimental results, Duflo et al. (2018) explores the value of regulatory
targeting in the state of Gujarat, India. The state has several major cities in violation of air quality standards,
and therefore the value of enforcement can be much larger. Yet in practice, inspections appear to be per-
formed less often than mandated. The paper implements an experimental treatment in which inspections to
polluting manufacturing plants are scheduled according to the regulation.54 However, the paper finds that
the reduced-form impacts of the experimental treatment are rather muted. To reconcile the findings, the au-
thors observe that regulators in the control group appear to be targeting their inspections. Even if regulators
inspect less often, they often appear to inspect the most suspect facilities. They build a dynamic model in
which regulators can use lagged values of pollution to target investment. They estimate the plant’s abate-
ment costs using a Rust-style estimator, taking the policy function of the regulator as given. Additionally,
they estimate the regulator’s preferences by assuming the observed policy is optimal. Using the model, they
confirm that increased random inspections are not particularly valuable in this setting.
Continuing to understand how to improve the monitoring and enforcement of environmental regulations
will be essential in the following decades, in which fossil fuel emissions need to decline dramatically. Work
on reliable and strategy-proof mechanisms to monitor emissions would be valuable. For example, more
work could be done in IO to study how to use satellite data to closely monitor worldwide emissions in a way
that can circumvent the major challenges of international cooperation and compliance. More work about
how to more reliably use satellite measurements will be highly valuable (see, for example, Torchiana et al.
(2020) in the context of land use changes in satellite data).
5.3 Regulation of water markets
The over-exploitation of environmental resources is another area that has gotten the attention of the energy
and environmental literature. IO tools are well-suited for the study of resource extraction problems, as54See also Duflo et al. (2013) for complementary work on the incentives faced by auditors, as this inspection treatment was
cross-randomized with an audit reform experiment, with auditors in the treatment being randomly assigned vs. chosen by theplants.
100
explained in section 2.1. In non-energy settings, water and fisheries are two areas in which IO tools have
been used to understand optimal extraction and evaluate or propose solutions for the allocation of these
scarce resources.
In this section, we focus on water markets, an area that is getting increasing attention. Water is a
resource that has been traditionally under-priced or even not priced at all. With the growing pressures on
water demand and the increased variance in rainfall, changes in the institutional treatment of water allocation
are likely to emerge in many areas (Libecap, 2011).
Water markets have historically been present in areas with substantial scarcity, such as Spain and Aus-
tralia, and are gaining prominence in the discussions surrounding other areas facing increasing scarcity.
Given the increasing pressures on earth’s natural resources, we expect a growing need for economics re-
searchers to help in the design and evaluation of emerging water markets.
5.3.1 Water use and adaptation
Timmins (2002) studies the pricing of groundwater in California in the 1990’s. He first confirms the basic
fact that water is vastly under-priced by comparing the marginal extraction costs of water (e.g., due to
pumping costs) with the price charged to consumers. Prices charged to consumers are always below cost in
his sample, and they tend to be much lower. Importantly, this cost comparison ignores the opportunity cost
of over-extracting water and depleting aquifers, along with the environmental costs of doing so. Timmins
(2002) develops a dynamic single-agent model to quantify the opportunity cost of water. The cost only
includes the increased pumping costs of declining aquifers, so it should be considered a lower bound. The
paper finds that the opportunity cost of depleting aquifers from a social point of view, yet, one observes
prices that are too low. To rationalize these observed patterns, the paper posits that water managers are
optimizing a welfare function with different weights for producer and consumer surplus. It then estimates
the model and finds that lower prices can be rationalized as the water manager catering to the constituents,
as opposed to minimizing costs.
From a methodological point of view, Timmins (2002) develops a single-agent model with persistent
unobserved shocks. His proposed approach uses a maximum likelihood approach to deal with the initial
conditions problem. The counterfactuals in Timmins (2002) assume that either managers internalized the
opportunity cost of water or not, which enables the quantification of the inefficient intertemporal allocation
of water due to myopic behavior. In addition to intertemporal mis-allocation, cross-sectional mis-allocation
of water is a concern in the absence of markets, an issue that is abstracted away in the paper.
Bruno and Jessoe (2021) and Burlig et al. (2021) study the demand for groundwater in California.55
Bruno and Jessoe (2021) uses data from the Coachella Valley Water District in the southern part of the state,
while Burlig et al. (2021) uses data covering all agricultural users in Pacific Gas & Electric’s service territory,
which includes most of California’s central valley. To estimate agricultural users’ demand elasticity, Bruno
and Jessoe (2021) uses variation in the price of groundwater itself, and Burlig et al. (2021) uses variation in
pumping costs induced by the fact that different farms are on different electricity tariff plans.55Bruno and Jessoe (2021) and Burlig et al. (2021) also include reviews of the literature estimating the elasticity of water demand.
101
Both Bruno and Jessoe (2021) and Burlig et al. (2021) use panel fixed effects regressions to identify the
water demand elasticity but arrive at different results. Bruno and Jessoe (2021) reports elasticity estimates
from -0.16 to -0.2, but the main estimate in Burlig et al. (2021) is -1.12. This large difference in results could
reflect differences in location (e.g. different types of crops grown in Northern versus Southern California)
or differences in the frequency and magnitude of the price changes in their data, conditional on the fixed
effects. Both papers use data at the monthly frequency, though the price changes observed in Burlig et al.
(2021) are long duration, and the paper finds similar elasticities when the data are collapsed to the annual
level. When Bruno and Jessoe (2021) uses annual data it estimates a larger elasticity of -0.37, though this
is still substantially smaller than in Burlig et al. (2021). Burlig et al. (2021) also finds evidence that crop
switching is the main mechanism behind farmers’ price responsiveness. Additional research is needed to
more fully understand the differences between these two papers’ results.
Hagerty (2017) focuses explicitly on long-run mis-allocation of surface water in California. The paper’s
regression approach takes advantage of differing approaches to the management of water across districts to
estimate the long-run responses in terms of revenue and crop choice to surface water scarcity. The paper
finds that crop revenues are significantly reduced even in the long-run, suggesting that adaptation cannot
mitigate many of the impacts of more scarce and volatile water availability.
5.3.2 Studying formal water markets
Several papers have also analyzed the functioning of institutionalized water markets, in which participants
have allowances to use water and can trade with others. We focus here on recent studies that use IO tools
for their analysis.
Rafey (2021) studies the southern Murray-Darling Basin (sMDB) in Australia during 2007-2015. The
paper proceeds in two steps. First, it uses production function tools (Olley and Pakes, 1996; Ackerberg et
al., 2015) to estimate unobserved productivity at agricultural plots. The estimation exploits detailed rotating
panel survey data together with meteorological data that together collect information on land, irrigation,
rainfall, and other flexible factors (labor and materials). Importantly, the approach combines the standard
strategy to control for drivers of productivity with lagged instruments based on the allocation rules for water
rights. Second, once productivity measures have been estimated, the paper uses detailed trading data to
examine the gains from trade in this market. The paper finds gains from trade in the order of 4-6%.
In a historical context, Donna and Espin-Sanchez (2018a) and Donna and Espin-Sanchez (2018b) study
the performance of water auctions in Mula (Spain) during the mid-twentieth century. Donna and Espin-
Sanchez (2018a) develops a model of auctions with complementarities to explore to what extent consecutive
blocks of irrigated water are complements or substitutes, finding that accounting for complementarities
in water use is important to obtain unbiased estimates. The results indicate complementarities are most
important in the driest summer months, consistent with the need to irrigate the dry canals for a longer time
before water can be productively used. In the same historical context, Donna and Espin-Sanchez (2018b)
compares an auction versus a quota mechanism to allocate water, exploiting a historical change in the way
water was allocated in Mula. The structural model takes into account market frictions, putting emphasis on
liquidity constraints faced by poorer farmers. In the presence of these liquidity constraints, the paper shows
102
that the elasticity of demand from a model ignoring these constraints is biased, overstating the elasticity
of demand, and that the quota system can be better as it does not preclude poorer farmers from cultivating
productive areas.
As stated above, we expect the work studying water markets to grow substantially in the next decade.
More work using IO tools to study the ex-ante ability and ex-post performance of markets to allocate water
in an efficient and equitable manner will be needed.
6 Concluding remarks
This chapter has discussed the variety of insights that industrial organization economists have made into
energy markets, market regulation, and environmental policy. These insights have stemmed from application
of nearly every tool in IO economists’ toolkit, ranging from models of differentiated product competition
to auction models to models of dynamic oligopoly. In many cases, IO economists’ work on energy and
environmental problems have enriched these models and their associated empirical methods in ways that are
applicable to economic and policy questions in other sectors.
Rather than recap the preceding pages, we conclude this paper by highlighting what we see as the most
important energy and environmental questions in need of future contributions from IO economists. Policy
interest in climate mitigation is especially intensifying, and insights from IO economists are needed to help
design and assess the likely impacts of policy options. In particular, we see the following areas as being
especially promising for future IO research:
• Productivity and innovation in energy technologies. Productivity growth and innovation have been
central in governing the world’s dominant sources of energy supply to date, and the future of zero-
emissions energy supply will arguably be governed more by the pace of innovation in clean technolo-
gies than by any other factor. Our chapter has highlighted the many contributions IO economists have
already made to understanding sources of productivity growth and innovation for both fossil and clean
energy sources. More work is needed to better understand factors that influence productivity growth,
how innovation is affected by market structure, and how policy can accelerate innovative activity in
green technologies.
• Industrial organization of zero-emissions energy sources. IO economists have delivered deep in-
sights into the economics of fossil fuel energy supply, and especially of the oil and gas industry.
Given the ascension of wind and solar power, and the possibilities for geothermal power and carbon
sequestration technologies in the future, there is an opportunity for IO economists to apply models
and tools from work on fossil fuels—for instance, models for leasing, auctions, productivity growth,
and investment dynamics—to these emerging green technologies.
• Electric and autonomous vehicles. EVs are beginning to achieve a non-trivial share of new vehicle
sales in many countries, so in the next few years we anticipate that a wealth of data on EV markets and
EV charging will become available. These new data will present IO economists with opportunities to
contribute to our understanding of EV markets and EV policy, well above and beyond the small set
103
of existing papers we have discussed in this chapter. Looking further ahead, the autonomous vehicle
(AV) industry is even more nascent than the EV industry, but as momentum gradually builds behind
autonomy, research will be needed to understand potential business models for AVs and the impacts of
policies proposed to govern this industry on consumers, energy use, and the environment. Ostrovsky
and Schwarz (2018) is an early theoretical contribution in this area that provides a framework for
jointly thinking about AVs, congestion pricing, and ride sharing. Additional work that further develops
this framework, with potential empirical applications, is needed.
• The future of electricity distribution. Improvements in technology are enabling households to invest
in distributed generation resources (especially rooftop solar), switch from natural gas to electricity for
home heating, and charge their EVs at home. Consumers’ willingness to choose these technologies
is likely to depend on electric utilities’ retail pricing policies. As noted in Borenstein and Kellogg
(2021), distribution utilities’ standard practice of incorporating their fixed costs into volumetric rates
will distort households’ incentives, since as the grid becomes cleaner retail prices are likely to sub-
stantially exceed social marginal cost. This standard practice may also lead to substantial inequities in
the presence of wide-scale adoption of distributed generation, which is likely to be disproportionately
taken up by high-income households (Borenstein and Davis, 2016). IO economists can contribute
to understanding the implications of potential alternative business models and regulatory frameworks
for electricity distribution and retailing that address both the efficiency and equity shortcomings of the
status quo.
• Integration of large-scale storage into wholesale electricity markets. The intermittency of wind
and solar energy resources implies that battery storage can potentially play an important role in cre-
ating an electricity grid that can supply zero-emission electricity on demand. It is not yet clear how
wholesale power markets will operate in the presence of such technologies and how they will provide
incentives for the investment in and use of battery storage. Input from IO economists will be vital in
evaluating the merits of alternative market designs, see for example Joskow (2019) and Wolak (2021).
• Stranded fossil fuel assets. In the event that climate policy leads to a rapid transformation of en-
ergy supply, fossil fuel physical and human capital are likely to be abandoned before the end of their
originally foreseen useful life. This premature abandonment will have substantial distributional impli-
cations, and it may also lead to environmental hazards in the event that physical capital is not properly
decommissioned. IO economists’ expertise in (dis)investment problems and utility regulation can help
inform policies aimed at accelerating asset decommissioning while equitably distributing the costs of
doing so.
References
Abito, Jose Miguel, “Measuring the Welfare Gains from Optimal Incentive Regulation,” The Review of
Economic Studies, 09 2019. rdz050.
104
Ackerberg, Daniel A., Kevin Caves, and Garth Frazer, “Identification Properties of Recent Production
Function Estimators,” Econometrica, 2015, 83 (6), 2411–2451.
Adams, William James and Janet L. Yellen, “Commodity Bundling and the Burden of Monopoly,” Quar-
terly Journal of Economics, 1976, 90 (3), 475 – 498.
Agerton, Mark, “Learning Where to Drill: Drilling Decisions and Geological Quality in the Haynesville
Shale,” 2020. working paper.
Aldy, Joseph E., Maximilian Auffhammer, Maureen Cropper, Arthur Fraas, and Richard Morgen-stern, “Looking Back at Fifty Years of the Clean Air Act,” Journal of Economic Literature, 2021.
Ale-Chilet, Jorge, Cuicui Chen, Jing Li, and Mathias Reynaert, “Colluding Against Environmental
Regulation,” 2021. working paper.
Allaz, Blaise and Jean-Luc Vila, “Cournot Competition, Forward Markets and Efficiency,” Journal of
Economic Theory, feb 1993, 59 (1), 1–16.
Allcott, Hunt and Nathan Wozny, “Gasoline Prices, Fuel Economy, and the Energy Paradox,” Review of
Economics and Statistics, 2014, 96 (5), 779 – 795.
Andersen, Laura Mørch, Lars Garn Hansen, Carsten Lynge Jensen, and Frank A. Wolak, “Can Incen-
tives to Increase Electricity Use Reduce the Cost of Integrating Renewable Resources,” National Bureau
of Economic Research, mar 2019.
Anderson, John W., Gordon W. Leslie, and Frank A. Wolak, “Measuring the Impact of Own and Others’
Experience on Project Costs in the U.S. Wind Generation Industry,” NBER working paper 26114 2019.
Anderson, Soren T. and James M. Sallee, “Using Loopholes to Reveal the Marginal Cost of Regulation:
The Case of Fuel-Economy Standards,” American Economic Review, 2011, 101 (4), 1375 – 1409.
and , “Designing Policies to Make Cars Greener: A Review of the Literature,” Annual Review of
Resource Economics, 2016, 8, 157 – 180.
, Ryan Kellogg, and James M. Sallee, “What Do Consumers Believe About Future Fuel Prices?,” Jour-
nal of Environmental Economics and Management, 2013, 66, 383 – 403.
, , and Stephen W. Salant, “Hotelling Under Pressure,” Journal of Political Economy, 2018, 126,
984–1026.
Arcidiacono, Peter and Robert A. Miller, “Conditional Choice Probability Estimation of Dynamic Dis-
crete Choice Models With Unobserved Heterogeneity,” Econometrica, nov 2011, 79 (6), 1823–1867.
Asker, John, Allan Collard-Wexler, and Jan De Loecker, “(Mis)Allocation, Market Power, and Global
Oil Extraction,” American Economic Review, 2019, 109 (4), 1568 – 1615.
105
Assuncao, Juliano, Robert Mcmillan, Joshua Murphy, Eduardo Souza-Rodrigues, Fernanda Brollo,Sylvain Chabe-Ferret, Francisco Costa, Clarissa Gandour, Kelsey Jack, Charles Manski, IsmaelMourifie, Marcel Oestreich, Martino Pelli, Alex Pfaff, Alberto Salvo, Paul Scott, Edson Severini,Kate Sims, and Aloysius Siow, “Optimal Environmental Targeting in the Amazon Rainforest *,” 2021.
Athey, Susan and Philip A. Haile, “Identification of Standard Auction Models,” Econometrica, 2002, 70
(6), 2107 – 2140.
Averch, Harvey and Leland L. Johnson, “Behavior of the Firm Under Regulatory Constraint,” The Amer-
ican Economic Review, 1962, 52 (5), 1052–1069.
Bajari, Patrick, Lanier Benkard, and Jonathan Levin, “Estimating Dynamic Models of Imperfect Com-
petition,” Econometrica, 2007, 75 (5), 1331 – 1370.
Baumeister, Christiane and James D. Hamilton, “Structural Interpretation of Vector Autoregressions with
Incomplete Identification: Revisiting the Role of Oil Supply and Demand Shocks,” American Economic
Review, 2019, 109 (5), 1873 – 1910.
Bento, Antonio M., Lawrence H. Goudler, Mark R. Jacobsen, and Roger H. von Haefen, “Distribu-
tional and Efficiency Impacts of Increased US Gasoline Taxes,” American Economic Review, 2009, 99
(3), 667 – 699.
Berry, Steven, James Levinsohn, and Ariel Pakes, “Automobile Prices in Market Equilibrium,” Econo-
metrica, 1995, 63 (4), 841 – 890.
Bhattacharya, Vivek, Andrey Ordin, and James W. Roberts, “Bidding and Drilling Under Uncertainty:
An Empirical Analysis of Contingent Payment Auctions,” 2020. working paper.
Birge, John R., Ali Hortacsu, Ignacia Mercadal, and J. Michael Pavlin, “Limits to arbitrage in electricity
markets: A case study of MISO,” Energy Economics, 2018.
Black, Fischer and Myron Scholes, “The Pricing of Options and Corporate Liabilities,” Journal of Political
Economy, 1973, 81 (3), 637 – 654.
Blonz, Joshua A, “Making the Best of the Second-Best: Welfare Consequences of Time-Varying Electricity
Prices,” Energy Institute at Haas Working Paper, 2016.
Blundell, Wesley, Gautam Gowrisankaran, and Ashley Langer, “Escalation of Scrutiny: The Gains from
Dynamic Enforcement of Environmental Regulations,” American Economic Review, forthcoming., 2020.
Board, Simon, “Selling Options,” Journal of Economic Theory, 2007, 136, 324 – 340.
Bollinger, Bryan K. and Wesley R. Hartmann, “Information vs. Automation and implications for dynamic
pricing,” Management Science, 2020.
Bolton, Patrick and Christopher Harris, “Strategic Experimentation,” Econometrica, 1999, 67 (2), 349 –
374.
106
Boomhower, Judson, “Drilling Like There’s No Tomorrow: Bankruptcy, Insurance, and Environmental
Risk,” American Economic Review, 2019, 109 (2), 391 – 426.
Borenstein, Severin, “The Private and Public Economics of Renewable Electricity Generation,” Journal of
Economic Perspectives, 2012, 26 (1), 67 – 92.
, “Private net benefits of residential solar pv: The role of electricity tariffs, tax incentives, and rebates,”
Journal of the Association of Environmental and Resource Economists, 2017.
and James Bushnell, “The US Electricity Industry after 20 Years of Restructuring,” Annual Review of
Economics, aug 2015, 7 (1), 437–463.
and Lucas W. Davis, “The Distributional Effects of U.S. Clean Energy Tax Credits,” NBER Tax Policy
and the Economy, 2016, 30 (1), 191–234.
and Ryan Kellogg, “Challenges of a Clean Energy Transition and Implications for Energy Infrastructure
Policy,” 2021. Aspen Economic Strategy Group paper.
, James B. Bushnell, and Frank A Wolak, “Measuring market inefficiencies in California’s restructured
wholesale electricity market,” American Economic Review, 2002, 92 (5), 1376–1405.
, James Bushnell, and Steven Stoft, “The Competitive Effects of Transmission Capacity in a Deregulated
Electricity Industry,” The RAND Journal of Economics, 2000, 31 (2), 294.
, , Christopher R Knittel, and Catherine Wolfram, “Inefficiencies and Market Power in Financial
Arbitrage: A Study of California’s Electricity Markets,” The Journal of Industrial Economics, 2008, 56
(2), 347–378.
BP, “Statistical Review of World Energy, 69th edition,” 2020. https://www.bp.com/content/
dam/bp/business-sites/en/global/corporate/pdfs/energy-economics/
statistical-review/bp-stats-review-2020-full-report.pdf.
Brehm, Paul A. and Eric Lewis, “To Trade or Not to Trade: Oil Leases, Information Asymmetry, and
Coase,” RAND Journal of Economics, forthcoming.
Bresnahan, Timothy F. and Peter C. Reiss, “Dealer and Manufacturer Margins,” RAND Journal of Eco-
nomics, 1985, 16 (2), 253 – 268.
Brown, David P. and Andrew Eckert, “Analyzing firm behavior in restructured electricity markets: Em-
pirical challenges with a residual demand analysis,” International Journal of Industrial Organization, jan
2021, 74, 102676.
Bruno, Ellen M. and Katrina Jessoe, “Missing markets: Evidence on agricultural groundwater demand
from volumetric pricing,” Journal of Public Economics, apr 2021, 196, 104374.
Burlig, Fiona, Louis Preonas, and Matt Woerman, “Energy, Groundwater, and Crop Choice,” 2021.
working paper.
107
Bushnell, J, E Mansur, and C Saravia, “Vertical Arrangments, Market Structure, and Competition: An
Analysis of Restructured US Electricity Markets,” American Economic Review, 2008, 98 (1), 237–266.
Bushnell, James and Kevin Novan, “Setting with the Sun: The Impacts of Renewable Energy on Wholesale
Power Markets,” National Bureau of Economic Research Working Paper Series, 2018.
Bushnell, Jim, Erin T. Mansur, and Kevin Novan, “Review of the Economics Literature on US Electricity
Restructuring,” 2017. UC Davis.
Busse, Meghan R. and Nathaniel O. Keohane, “Market Effects of Environmental Regulation: Coal, Rail-
roads, and the 1990 Clean Air Act,” RAND Journal of Economics, 2008, 38 (4), 1159 – 1179.
, Christopher R. Knittel, and Florian Zettelmeyer, “Are Consumers Myopic? Evidence from New and
Used Car Purchases,” American Economic Review, 2013, 103 (1), 220 – 256.
Butters, R. Andrew, Jackson Dorsey, and Gautam Gowrisankaran, “Soaking Up the Sun: Battery In-
vestment, Renewable Energy, and Market Equilibrium,” Technical Report aug 2020.
Byrne, David P. and Nicolas de Roos, “Consumer Search in Retail Gasoline Markets,” Journal of Industrial
Economics, 2017, 65 (1), 183 – 193.
, Leslie A. Martin, and Jia Sheen Nah, “Price Discrimination, Search, and Negotiation in an Oligopoly:
A Field Experiment in Retail Electricity,” SSRN Electronic Journal, 2019.
Callaway, Duncan S., Meredith Fowlie, and Gavin McCormick, “Location, location, location: The vari-
able value of renewable energy and demand-side efficiency resources,” Journal of the Association of
Environmental and Resource Economists, 2018.
Carlson, Curtis, Dallas Burtraw, Maureen Cropper, and K. Palmer, “Sulfur dioxide control by electric
utilities: What are the gains from trade?,” Journal of Political Economy, jul 2000, 108 (6), 1292–1326.
Chandra, Ambarish and Mariano Tappata, “Consumer Search and Dynamic Price Dispersion: an Appli-
cation to Gasoline Markets,” RAND Journal of Economics, 2011, 42 (4), 681 – 704.
Chen, Cuicui, “Identifying Beliefs in Continuous-Action Dynamic Models: An Application to the US SO2
Allowance Market,” Mimeo, 2021.
Cherniwchan, Jevan, Brian R. Copeland, and M. Scott Taylor, “Trade and the Environment: New Meth-
ods, Measurements, and Results,” https://doi.org/10.1146/annurev-economics-063016-103756, aug 2017,
9, 59–85.
Cicala, Steve, “When Does Regulation Distort Costs? Lessons from Fuel Procurement in US Electricity
Generation,” American Economic Review, January 2015, 105 (1), 411–44.
, “Imperfect Markets versus Imperfect Regulation in U.S. Electricity Generation,” American Economic
Review, 2020.
108
Clay, Karen, Akshaya Jha, Nicholas Muller, , and Randall Walsh, “External Costs of Transporting
Petroleum Products: Evidence from Shipments of Crude OIl from North Dakota by Pipelines and Rail,”
Energy Journal, 2019, 40 (1), 55 – 72.
Coase, R. H., “The Problem of Social Cost,” The Journal of Law and Economics, 1960, 3, 1–44.
Cole, Matthew T., Jacqueline M. Doremus, and Stephen F. Hamilton, “Import Restrictions by Eco-
Certification: Quantity Effects on Tropical Timber Production,” Journal of Environmental Economics
and Management, 2021, 107, 102423.
Copeland, Brian R. and M. Scott Taylor, “Trade, Growth, and the Environment,” Journal of Economic
Literature, 3 2004, 42, 7–71.
Covert, Thomas R., “Experiential and Social Learning in Firms: The Case of Hydraulic Fracturing in the
Bakken Shale,” 2015. working paper.
and Richard L. Sweeney, “Relinquishing Riches: Auctions vs Informal Negotiations in Texas Oil and
Gas Leasing,” NBER working paper 25712 2019.
and Ryan Kellogg, “Crude by Rail, Option Value, and Pipeline Investment,” NBER working paper
23855 2018.
Cullen, Joseph, “Measuring the environmental benefits of wind-generated electricity,” American Economic
Journal: Economic Policy, 2013, 5 (4), 107–133.
, “Dynamic Response to Environmental Regulation in the Electricity Industry,” Working Paper, 2015.
Cullen, Joseph A. and Erin T. Mansur, “Inferring carbon abatement costs in electricity markets: A re-
vealed preference approach using the shale revolution,” American Economic Journal: Economic Policy,
2017.
Cullen, Joseph and Stanley Reynolds, “Market Dynamics and Investment in the Electricity Sector,” Work-
ing Paper, 2015.
Currie, Janet and Reed Walker, “What Do Economists Have to Say about the Clean Air Act 50 Years after
the Establishment of the Environmental Protection Agency?,” Journal of Economic Perspectives, 2019,
33 (4), 3–26.
Dardati, Evangelina, “Pollution permit systems and firm dynamics: How does the allocation scheme mat-
ter?*,” International Economic Review, feb 2016, 57 (1), 305–328.
Davis, Lucas and Catherine Hausman, “Market impacts of a nuclear power plant closure,” American
Economic Journal: Applied Economics, 2016, 8 (2), 92–122.
Davis, Lucas W and Catherine Hausman, “Who will pay for legacy utility costs?,” 2021.
109
Davis, Lucas W. and Catherine Wolfram, “Deregulation, Consolidation, and Efficiency: Evidence from
US Nuclear Power,” American Economic Journal: Applied Economics, July 2012, 4 (4), 194–225.
De Loecker, Jan and Frederic Warzynski, “Markups and Firm-Level Export Status,” American Economic
Review, oct 2012, 102 (6), 2437–71.
DeMarzo, Peter M., Ilan Kremer, and Andrzej Skrzypacz, “Bidding with Securities: Auctions and Se-
curity Design,” American Economic Review, 2005, 95 (4), 936 – 959.
Deryugina, Tatyana, Alexander MacKay, and Julian Reif, “The long-run dynamics of electricity de-
mand: Evidence from municipal aggregation,” American Economic Journal: Applied Economics, 2020.
Dixit, Avinash K. and Robert S. Pindyck, Investment Under Uncertainty, Princeton, NJ:: Princeton Uni-
versity Press, 1994.
Donna, Javier and Jose-Antonio Espin-Sanchez, “Complements and Substitutes in Sequential Auctions:
The Case of Water Auctions,” RAND Journal of Economics, feb 2018, 49 (1).
and , “The Illiquidity of Water Markets: Efficient Institutions for Water Allocation in Southeastern
Spain,” SSRN Electronic Journal, aug 2018.
Dorsey, Jackson, Ashley Langer, and Shaun McRae, “Fueling Alternatives: Gas Station Choice, the
Value of Time, and Implications for Electric Charging,” 2021. working paper.
Dorsey-Palmateer, Reid, “Effects of Wind Power Intermittency on Generation and Emissions,” The Elec-
tricity Journal, 2019, 32 (3), 25–30.
Duflo, Esther, Michael Greenstone, Rohini Pande, and Nicholas Ryan, “Truth-telling by third-party
auditors and the response of polluting firms: Experimental evidence from india,” Quarterly Journal of
Economics, 2013.
, , , and , “The Value of Regulatory Discretion: Estimates From Environmental Inspections in
India,” Econometrica, 2018.
Duso, Tomaso and Florian Szucs, “Market power and heterogeneous pass-through in German electricity
retail,” European Economic Review, 2017, 98 (C), 354–372.
Ederington, Josh, Arik Levinson, and Jenny Minier, “Footloose and pollution-free,” Review of Eco-
nomics and Statistics, 2005, 87 (1), 92–99.
Eisenberg, Thomas, “Regulatory Distortions and Capacity Investment: The Case of China’s Coal Power
Industry,” Mimeo, 2020.
Ellerman, A. Denny and Juan Pablo Montero, “The efficiency and robustness of allowance banking in
the U.S. Acid Rain Program,” Energy Journal, 2007, 28 (4), 47–71.
110
Environmental Protection Agency, “Inventory of U.S. Greenhouse Gas Emissions and Sinks: 1990–
2019,” 2021. Washington, D.C. https://www.epa.gov/sites/production/files/2021-02/documents/us-ghg-
inventory-2021-main-text.pdf.
Fabra, N. and M. Reguant, “Pass-through of emissions costs in electricity markets,” American Economic
Review, 2014, 104 (9).
Fabra, Natalia, David Rapson, Mar Reguant, and Jingyuan Wang, “Estimating the Elasticity to Real
Time Pricing: Evidence from the Spanish Electricity Market,” American Economic Review Papers and
Proceedings, 2021.
Fabrizio, Kira R., Nancy L. Rose, and Catherine D. Wolfram, “Do Markets Reduce Costs? Assessing the
Impact of Regulatory Restructuring on US Electric Generation Efficiency,” American Economic Review,
September 2007, 97 (4), 1250–1277.
Fan, Ying, “Ownership Consolidation and Product Characteristics: A Study of the US Daily Newspaper
Market,” American Economic Review, 2013, 103 (5), 1598 – 1628.
Feger, Fabian, Nicola Pavanini, and Doina Radulescu, “Welfare and Redistribution in Residential Elec-
tricity Markets with Solar Power,” Working paper, 2017.
Fell, Harrison, Daniel T. Kaffine, and Kevin Novan, “Emissions, Transmission, and the Environmental
Value of Renewable Energy,” American Economic Journal: Economic Policy, 2021.
Fetter, T. Robert, Andrew L. Steck, Christopher Timmins, and Douglas H. Wrenn, “Learning by View-
ing? Social Learning, Regulatory Disclosure, and Firm Productivity in Shale Gas,” 2020. working paper.
Fowlie, M., M. Reguant, and S.P. Ryan, “Market-based emissions regulation and industry dynamics,”
Journal of Political Economy, 2016, 124 (1).
Fowlie, Meredith, “Emissions Trading, Electricity Restructuring, and Investment in Pollution Abatement,”
American Economic Review, June 2010, 100 (3), 837–69.
and Jeffrey M. Perloff, “Distributing Pollution Rights in Cap-and-Trade Programs: Are Outcomes
Independent of Allocation?,” Review of Economics and Statistics, dec 2013, 95 (5), 1640–1652.
and Mar Reguant, “Challenges in the Measurement of Leakage Risk,” AEA Papers and Proceedings,
may 2018, 108, 124–129.
and Nicholas Muller, “Market-Based Emissions Regulation When Damages Vary across Sources:
What Are the Gains from Differentiation?,” Journal of the Association of Environmental and Resource
Economists, 2019, 6 (3), 593–632.
, Catherine Wolfram, Annika Todd, Patrick Baylis, and Peter Cappers, “Default Effects and Follow-
On Behavior: Evidence from an Electricity Pricing Program,” Review of Economic Studies, 2020.
111
Fowlie, Meredith L. and Mar Reguant, “Mitigating Emissions Leakage in Incomplete Carbon Markets,”
Journal of the Association of Environmental and Resource Economists, 2021, forthcomin.
Fox, Jeremy T., Kyoo il Kim, Stephen P. Ryan, and Patrick Bajari, “A simple estimator for the distribu-
tion of random coefficients,” Quantitative Economics, 2011.
Ganapati, Sharat, Joseph S. Shapiro, and Reed Walker, “Energy Cost Pass-Through in US Manufac-
turing: Estimates and Implications for Carbon Taxes,” American Economic Journal: Applied Economics,
apr 2020, 12 (2), 303–42.
Gerarden, Todd, “Demanding Innovation: The Impact of Consumer Subsidies on Solar Panel Production
Costs,” Technical Report, Cornell University 2019.
Gillingham, Kenneth T., Sebastien Houde, and Arthur A. Van Benthem, “Consumer Myopia in Vehi-
cle Purchases: Evidence from a Natural Experiment,” American Economic Journal: Economic Policy,
forthcoming.
Goettler, Ronald L. and Brett R. Gordon, “Does AMD Spur Intel to Innovate More?,” Journal of Political
Economy, 2011, 119, 1141 – 1200.
Goldberg, Pinelopi Koujianou, “The Effects of the Corporate Average Fuel Efficiency Standards in the
US,” Journal of Industrial Economics, 1998, 46 (1), 1 – 33.
Gowrisankaran, Gautam, Stanley S Reynolds, and Mario Samano, “Intermittency and the value of
renewable energy,” Journal of Political Economy, 2015, pp. 1–52.
Green, R. J. and D. M. Newbery, “Competition in the British electricity spot market,” Journal of Political
Economy, 1992.
Grigolon, Laura, Mathias Reynaert, and Frank Verboven, “Consumer Valuation of Fuel Costs and Tax
Policy: Evidence from the European Car Market,” American Economic Journal: Economic Policy, 2018,
10 (3), 193 – 225.
Groote, Olivier De and Frank Verboven, “Subsidies and Time Discounting in New Technology Adoption:
Evidence from Solar Photovoltaic Systems,” American Economic Review, June 2019, 109 (6), 2137–72.
Hagerty, Nick, “Liquid Constrained in California: Estimating the Potential Gains from Water Markets,”
Unpublished Ph.D thesis, 2017, pp. 1–75.
Hahn, Robert W. and Robert N. Stavins, “The Effect of Allowance Allocations on Cap-and-Trade System
Performance,” The Journal of Law and Economics, nov 2011, 54 (S4), S267–S294.
Haile, Philip, Kenneth Hendricks, and Robert Porter, “Recent U.S. Offshore Oil and Gas Lease Bidding:
A Progress Report,” International Journal of Industrial Organization, 2010, 28, 390–396.
Hall, Robert, “Market Structure and Macroeconomic Fluctuations,” Brookings Papers on Economic Activ-
ity, 1986, 17 (2), 285–338.
112
Hamilton, James D., “Historical Oil Shocks,” in Randall E. Parker and Robert Whaples, eds., Routledge
Handbook of Major Events in Economic History, New York: Routledge Taylor and Francis Group, 2013.
Harding, Matthew and Steven Sexton, “Household Response to Time-Varying Electricity Prices,” Annual
Review of Economics, October 2017, 9 (1), 337–359.
Hausman, Catherine, “Corporate Incentives and Nuclear Safety,” American Economic Journal: Economic
Policy, 2014, 6, 178–206.
Heckman, James J., “Heterogeneity and State Dependence,” in Sherwin Rosen, ed., Studies in Labor
Markets, University of Chicago Press, 1981.
Hendricks, Ken and Robert H. Porter, “An Empirical Perspective on Auctions,” in M. Armstrong and
R. Porter, eds., Handbook of Industrial Organization, North Holland, 2007.
, , and Bryan Boudreau, “Information, Returns, and Bidding Behavior in Ocs Auctions, 1954–1969,”
Journal of Industrial Economics, 1987, 35 (4), 517 – 542.
, , and Guofo Tan, “Optimal Selling Strategies for Oil and Gas Leases with an Informed Buyer,”
American Economic Review Papers and Proceedings, 1993, 83 (2), 234–239.
Hendricks, Kenneth and Dan Kovenock, “Asymmetric Information, Information Externalities, and Effi-
ciency: the Case of Oil Exploration,” RAND Journal of Economics, 1989, 20 (2), 164 – 182.
and Robert H. Porter, “The Timing and Incidence of Exploratory Drilling on Offshore Wildcat Tracts,”
American Economic Review, 1996, 86 (3), 388 – 407.
Herfindahl, Orris C., “Depletion and Economic Theory,” in Mason Gaffney, ed., Extractive Resources and
Taxation, University of Wisconsin Press, 1967.
Herrnstadt, Evan M., Ryan Kellogg, and Eric Lewis, “The Economics of Time-Limited Development
Options: The Case of Oil and Gas Leases,” NBER working paper 27165 2020.
Ho, Jacqueline S., Jhih-Shyang Shih, Lucija Muehlenbachs, Clayton Munnings, and Alan J. Krup-nick, “Managing Environmental Liability: An Evaluation of Bonding Requirements for Oil and Gas
Wells in the United States,” Environmental Science and Technology, 2018, 52, 3908 – 3916.
Hodgson, Charles, “Information Externalities, Free Riding, and Optimal Exploration in the UK Oil Indus-
try,” 2018. working paper.
Hopenhayn, Hugo A., “Entry, Exit, and firm Dynamics in Long Run Equilibrium,” Econometrica, 1992.
Hortacsu, Ali and Steven L. Puller, “Understanding strategic bidding in multi-unit auctions: A case study
of the Texas electricity spot market,” RAND Journal of Economics, 2008, 39 (1), 86–114.
, Fernando Luco, Steven L. Puller, and Dongni Zhu, “Does strategic ability affect efficiency? Evidence
from electricity markets,” American Economic Review, 2019.
113
, Seyed Ali Madanizadeh, and Steven L. Puller, “Power to choose? An analysis of consumer inertia in
the residential electricity market,” American Economic Journal: Economic Policy, 2017.
Hotelling, Harold, “The Economics of Exhaustible Resources,” Journal of Political Economy, 1931, 39 (2),
137 – 175.
Hsiao, Allan, “Coordination and Commitment in International Climate Action: Evidence from Palm Oil,”
Working Paper, 2020.
Hughes, Jonathan E., “The Higher Price of Cleaner Fuels: Market Power in the Rail Transport of Fuel
Ethanol,” Journal of Environmental Economics and Management, 2011, 62 (2), 123 – 139.
and Ian Lange, “Who (Else) Benefits from Electricity Deregulation? Coal Prices, Natural Gas, and
Price Discrimination,” Economic Inquiry, 2020, 58 (3), 1053 – 1075.
Hyne, Norman J., Nontechnical Guide to Petroleum Geology, Exploration, Drilling, and Production, 2nd
ed., Tulsa, Oklahoma: PennWell Corporation, 2001.
Ifrach, Bar and Gabriel Y. Weintraub, “A Framework for Dynamic Oligopoly in Concentrated Indus-
tries,” The Review of Economic Studies, 09 2016, 84 (3), 1106–1150.
Ito, K. and M. Reguant, “Sequential markets, market power, and arbitrage,” American Economic Review,
2016, 106 (7), 1921–1957.
Ito, Koichiro, “Do consumers respond to marginal or average price? Evidence from nonlinear electricity
pricing,” American Economic Review, 2014, 104 (2), 537–563.
and James M. Sallee, “The Economics of Attribute-Based Regulation: Theory and Evidence from Fuel
Economy Standards,” Review of Economics and Statistics, 2018, 100 (2), 319 – 336.
, Takanori Ida, and Makoto Tanaka, “Moral Suasion and economic incentives: Field experimental
evidence from energy demand,” American Economic Journal: Economic Policy, 2018.
Jacobsen, Mark R., “Evaluating U.S. Fuel Economy Standards in a Model with Producer and Household
Heterogeneity,” American Economic Journal: Economic Policy, 2013, 5 (2), 148 – 187.
and Arthur A. van Benthem, “Vehicle Scrappage and Gasoline Policy,” American Economic Review,
2015, 105 (3), 1312 – 1338.
Jessoe, Katrina and David Rapson, “Knowledge is (Less) power: Experimental evidence from residential
energy use,” American Economic Review, 2014.
and , “Commercial and Industrial Demand Response Under Mandatory Time-of-Use Electricity Pric-
ing,” Journal of Industrial Economics, 2015.
Jha, Akshaya and Frank A. Wolak, “Can Forward Commodity Markets Improve Spot Market Perfor-
mance? Evidence from Wholesale Electricity,” SSRN Electronic Journal, 2020.
114
Joskow, Paul and Jean Tirole, “Reliability and competitive electricity markets,” RAND Journal of Eco-
nomics, 2007.
Joskow, Paul L., “Comparing the Costs of Intermittent and Dispatchable Electricity Generating Technolo-
gies,” American Economic Review, 2011, 101 (3), 238 – 41.
, “Challenges for wholesale electricity markets with intermittent renewable generation at scale: The US
experience,” Oxford Review of Economic Policy, 2019.
and Jean Tirole, “Transmission Rights and Market Power on Electric Power Networks,” The RAND
Journal of Economics, 2000, 31 (3), 450.
Kaffine, Daniel T., Brannin J. McBee, and Jozef Lieskovsky, “Emissions savings from wind power gen-
eration in Texas,” Energy Journal, 2013.
Kalouptsidi, Myrto, Paul T. Scott, and Eduardo Souza-Rodrigues, “Linear IV regression estimators for
structural dynamic discrete choice models,” Journal of Econometrics, may 2021, 222 (1), 778–804.
Kang, Karam and Bernardo S Silveira, “Understanding Disparities in Punishment: Regulator Preferences
and Expertise,” Working Paper, 2018.
Karaduman, Omer, “Economics of Grid-Scale Energy Storage,” Mimeo, 2020.
Kellogg, Ryan, “Learning by Drilling: Inter-Firm Learning and Relationship Persistence in the Texas Oil-
patch,” Quarterly Journal of Economics, 2011, 126 (4), 1961 – 2004.
, “The Effect of Uncertainty on Investment: Evidence from Texas Oil Drilling,” American Economic
Review, 2014, 104 (6), 1698 – 1734.
Kilian, Lutz and Robert J. Vigfusson, “The Role of Oil Price Shocks in Causing U.S. Recessions,” Journal
of Money, Credit, and Banking, 2017, 49 (8), 1747 – 1776.
Kim, Harim, “Heterogeneous Impacts of Cost Shocks, Strategic Bidding, and Pass-Through: Evidence
from the New England Electricity Market,” American Economic Journal: Microeconomics, 2021.
Klemperer, Paul D. and Margaret A. Meyer, “Supply Function Equilibria in Oligopoly under Uncer-
tainty,” Econometrica, 1989.
Kleven, Henrik Jacobsen, “Bunching,” Annual Review of Economics, 2016, 8, 435 – 464.
Klier, Thomas and Joshua Linn, “New-vehicle Characteristics and the Cost of the Corporate Average Fuel
Economy Standard,” RAND Journal of Economics, 2012, 43 (1), 186 – 213.
Knittel, Christopher R., “Automobiles on Steroids: Product Attribute Trade-Offs and Technological
Progress in the Automobile Sector,” American Economic Review, 2011, 101 (7), 3368 – 3399.
115
Kolstad, Jonathan and Frank Wolak, “Using Environmental Emissions Permit Prices to Raise Electricity
Prices : Evidence from the California Electricity Market,” Center for the Study of Energy Markets, 2008,
(August), 1–62.
Kong, Yunmi, “Not Knowing the Competition: Evidence and Implications for Auction Design,” RAND
Journal of Economics, 2020, 51 (3), 840 – 867.
, “Sequential Auctions with Synergy and Affiliation Across Auctions,” Journal of Political Economy,
forthcoming.
, Isabelle Perrigne, and Quang Vuong, “Multidimensional Auctions of Contracts: An Empirical Anal-
ysis,” 2020. working paper.
Krautkraemer, Jeffrey A., “Nonrenewable Resource Scarcity,” Journal of Economic Literature, 1998, 36
(4), 2065 – 2107.
Kwoka, John, “Restructuring the U.S. Electric Power Sector: A Review of Recent Studies,” Review of
Industrial Organization, 2008, 32 (3/4), 165–196.
Laffont, Jean-Jacques and Jean Tirole, “Using Cost Observation to Regulate Firms,” Journal of Political
Economy, 1986, 94 (3), 614 – 641.
and , A Theory of Incentives in Procurement and Regulation number 0262121743. In ‘MIT Press
Books.’, The MIT Press, August 1993.
Langer, Ashley and Derek Lemoine, “Designing Dynamic Subsidies to Spur Adoption of New Technolo-
gies,” Technical Report, University of Arizona 2018.
Levinson, Arik and M. Scott Taylor, “Unmasking the pollution haven effect,” International Economic
Review, feb 2008, 49 (1), 223–254.
Levitt, Clinton J., “Information Spillovers in Onshore Oil and Gas Exploration,” Resource and Energy
Economics, 2016, 45, 80 – 98.
Lewis, Eric, “Patchwork Policies, Spillovers, and the Search for Oil and Gas,” American Economic Journal:
Economic Policy, 2019, 11 (1), 380 – 405.
Lewis, Matthew S. and Howard P. Marvel, “When Do Consumer Search?,” Journal of Industrial Eco-
nomics, 2011, 59 (3), 457 – 483.
Li, Jing, “Compatibility and Investment in the U.S. Electric Vehicle Market,” 2019. working paper.
Li, Shanjun, Lang Ton, Jianwei Xing, and Yiyi Zhou, “The Market for Electric Vehicles: Indirect Net-
work Effects and Policy Design,” Journal of the Association of Environmental and Resource Economists,
2017, 4 (1), 89 – 133.
116
Libecap, Gary D., “Institutional Path Dependence in Climate Adaptation: Coman’s “Some Unsettled Prob-
lems of Irrigation”,” American Economic Review, feb 2011, 101 (1), 64–80.
Lim, Claire S. H. and Ali Yurukoglu, “Dynamic Natural Monopoly Regulation: Time Inconsistency,
Moral Hazard, and Political Environments,” Journal of Political Economy, 2018, 126 (1), 263–312.
Lin, Cynthia C.-Y., “Strategic Decision-Making with Information and Extraction Externalities: a Struc-
tural Model of the Multistage Investment Timing Game in Offshore Petroleum Production,” Review of
Economics and Statistics, 2013, 95 (5), 1601 – 1621.
Linn, Joshua and Kristen McCormack, “The roles of energy markets and environmental regulation in
reducing coal-fired plant profits and electricity sector emissions,” RAND Journal of Economics, 2019.
Lyubich, Eva, Joseph Shapiro, and Reed Walker, “Regulating Mismeasured Pollution: Implications of
Firm Heterogeneity for Environmental Policy,” AEA Papers and Proceedings, may 2018, 108, 136–42.
Magnac, Thierry and David Thesmar, “Identifying dynamic discrete decision processes,” Econometrica,
2002.
Mahadevan, Meera, “The Price of Power: Costs of Political Corruption in Indian Electricity,” 2021. work-
ing paper.
Malik, Kabir, Maureen Cropper, Alexander Limonov, and Anoop Singh, “The Impact of Electricity
Sector Restructuring on Coal-fired Power Plants in India,” The Energy Journal, 2015, 36 (4), 287–312.
Mansur, Erin T, “Measuring Welfare in Restructured Electricity Markets,” Review of Economics and Statis-
tics, 2008, 90 (2), 369–386.
Mansur, Erin T. and Matthew W. White, “Market Organization and Efficiency in Electricity Markets,”
Mimeo, 2012, pp. 1–56.
Maskin, Eric and Jean Tirole, “A Theory of Dynamic Oligopoly II: Price Competition, Kinked Demand
Curves, and Edgeworth Cycles,” Econometrica, 1988, 56 (3), 571 – 599.
Mercadal, Ignacia, “Dynamic competition and arbitrage in electricity markets: The role of financial play-
ers,” American Economic Journal: Microeconomics, 2021.
Milgrom, Paul R. and Robert J. Weber, “A Theory of Auctions and Competitive Bidding,” Econometrica,
1982, 50 (5), 1089–1122.
Miravete, Eugenio J., Marıa J. Moral, and Jeff Thurk, “Fuel Taxation, Emissions Policy, and Compet-
itive Advantage in the Diffusion of European Diesel Automobiles,” RAND Journal of Economics, 2018,
49 (3), 504 – 540.
Muehlegger, Erich and Richard L. Sweeney, “Pass-Through of Own and Rival Cost Shocks: Evidence
from the U.S. Fracking Boom,” National Bureau of Economic Research, nov 2017.
117
Muehlenbachs, Lucija, “A Dynamic Model of Cleanup: Estimating Sunk Costs in Oil and Gas Production,”
International Economic Review, 2015, 56 (1), 155 – 185.
Muller, Nicholas Z. and Robert Mendelsohn, “Efficient Pollution Regulation: Getting the Prices Right,”
American Economic Review, dec 2009, 99 (5), 1714–39.
Nemet, Gregory F., “Beyond the learning curve: factors influencing cost reductions in photovoltaics,”
Energy Policy, 2006.
, “Subsidies for New Technologies and Knowledge Spillovers from Learning by Doing,” Journal of Policy
Analysis and Management, 2012.
Newell, Richard G. and Brian C. Prest, “The Unconventional Oil Supply Boom: Aggregate Price Re-
sponse from Microdata,” Energy Journal, 2019, 40 (3), 1 – 30.
, , and Ashley B. Vissing, “Trophy Hunting versus Manufacuring Energy: The Price Responsiveness
of Shale Gas,” Journal of the Association of Environmental and Resource Economists, 2019, 6 (2), 391 –
431.
Nguyen, Anh, “Information Design in Common Value Auction with Moral Hazard: Application to OCS
Leasing Auctions,” 2021. working paper.
Novan, Kevin, “Valuing the Wind: Renewable Energy Policies and Air Pollution Avoided,” American Eco-
nomic Journal: Economic Policy, 2015, 7 (3), 291–326.
Oliva, Paulina, “Environmental regulations and corruption: Automobile emissions in Mexico city,” Journal
of Political Economy, oct 2015, 123 (3), 686–724.
Olley, G. Steven and Ariel Pakes, “The Dynamics of Productivity in the Telecommunications Equipment
Industry,” Econometrica, nov 1996, 64 (6), 1263.
O’Neill, Richard P., Paul M. Sotkiewicz, Benjamin F. Hobbs, Michael H. Rothkopf, and William R.Stewart, “Efficient market-clearing prices in markets with nonconvexities,” European Journal of Opera-
tional Research, July 2005, 164 (1), 269–285.
Ordin, Andrey, “Investment and Taxation: the Case of Oil and Gas in the Permian Basin,” 2019. working
paper.
Ostrovsky, Michael and Michael Schwarz, “Carpooling and the Economics of Self-Driving Cars,” 2018.
working paper.
Pakes, Ariel, Michael Ostrovsky, and Steven Berry, “Simple Estimators for the Parameters of Discrete
Dynamic Games (with Entry/Exit Examples),” RAND Journal of Economics, 2007, 38, 373 – 399.
Patel, April and Emily Geary, “U.S. crude oil and natural gas production in 2019 hit records with fewer
rigs and wells,” EIA Today In Energy, June 25 2020.
118
Pindyck, Robert S., “The Optimal Exploration and Production of Nonrenewable Resources,” Journal of
Political Economy, 1978, 86 (5), 841 – 861.
, “The Dynamics of Commodity Spot and Futures Markets: A Primer,” Energy Journal, 2001, 22 (3), 1 –
29.
Porter, Robert H., “The Role of Information in U.S. Offshore Oil and Gas Lease Auction,” Econometrica,
1995, 63 (1), 1 – 27.
Preonas, Louis, “Market Power in Coal Shipping and Implications for U.S. Climate Policy,” 2019. working
paper.
Puller, Steven L., “Pricing and Firm Conduct in California’s Deregulated Electricity Market,” Review of
Economics and Statistics, 2007, 89 (1).
Qiu, Chenyu, “The Value of Wind Information in Wholesale Electricity Market: Evidence from U.S. Mid-
west,” 2020. working paper.
Rafey, Will, “Droughts, deluges, and (river) diversions: Valuing market-based water reallocation,” Mimeo,
2021.
Raimi, Daniel, Alan Krupnick, Jhih-Shyang Shah, and Alexandra Thompson, “Decommissioning Or-
phaned and Abandoned Oil and Gas Wells: New Estimates and Cost Drivers,” 2021. preprint.
Reguant, M., “Complementary bidding mechanisms and startup costs in electricity markets,” Review of
Economic Studies, 2014, 81 (4).
Reguant, Mar and A. D Ellerman, “Grandfathering and the endowment effect An Assessment in the
context of the Spanish National Allocation Plan,” Center for Energy and Environmental Policy Research,
2008, 49, 69–73.
Reiss, Peter C. and Matthew W. White, “Household electricity demand, revisited,” Review of Economic
Studies, 2005, 72 (3), 853–883.
Reynaert, Mathias, “Abatement Strategies and the Cost of Environmental Regulation: Emission Standards
on the European Car Market,” Review of Economic Studies, forthcoming.
and James M. Sallee, “Who Benefits When Firms Game Corrective Policies?,” American Economic
Journal: Economic Policy, forthcoming.
Riley, John G., “Ex Post Information in Auctions,” Review of Economic Studies, 1988, 55 (3), 409 – 429.
Rust, John, “Optimal Replacement of GMC Bus Engines: An Empirical Model of Harold Zurcher,” Econo-
metrica, 1987, 55 (5), 999 – 1033.
Ryan, Nicholas, “The Competitive Effects of Transmission Infrastructure in the Indian Electricity Market,”
American Economic Journal: Microeconomics, 2021.
119
Ryan, Stephen P., “The Costs of Environmental Regulation in a Concentrated Industry,” Econometrica,
2012, 80 (3), 1019–1061.
Sallee, James M., Sarah E. West, and Wei Fan, “Do Consumers Recognize the Value of Fuel Economy?
Evidence from Used Car Prices and Gasoline Price Fluctuations,” Journal of Public Economics, 2016,
135, 61 – 73.
Salop, Steven C., “Monopolistic Competition with Outside Goods,” Bell Journal of Economics, 1979, 10,
141 – 156.
Schmalensee, Richard and Robert N. Stavins, “The SO2 allowance trading system: The ironic history of
a grand policy experiment,” Journal of Economic Perspectives, dec 2013, 27 (1), 103–122.
, Paul L. Joskow, A. Denny Ellerman, Juan Pablo Montero, and Elizabeth M. Bailey, “An Interim
Evaluation of Sulfur Dioxide Emissions Trading,” Journal of Economic Perspectives, 1998, 12 (3), 53–68.
Scott, Paul T, “Dynamic Discrete Choice Estimation of Agricultural Land Use,” 2013.
Shaffer, Blake, “Misunderstanding nonlinear prices: Evidence from a natural experiment in electricity
demand,” American Economic Journal: Economic Policy, 2020, 12 (3).
Shavell, Steven, “The Judgment Proof Problem,” International Review of Law and Economics, 1986, 6 (1),
45 – 58.
Skrzypacz, Andrzej, “Auctions with Contingent Payments—an Overview,” International Journal of Indus-
trial Organization, 2013, 31, 666 – 675.
Slade, Margaret E. and Henry Thille, “Whither Hotelling: Tests of the Theory of Exhaustible Resources,”
Annual Review of Resource Economics, 2009, 1 (1), 239 – 259.
Small, Kenneth A., “Valuation of Travel Time,” Economics of Transportation, 2012, 1, 2 – 14.
Souza-Rodrigues, Eduardo, “Deforestation in the Amazon: A Unified Framework for Estimation and
Policy Analysis,” The Review of Economic Studies, 11 2019, 86, 2713–2744.
Springel, Katalin, “Network Externality and Subsidy Structure in Two-Sides Markets: Evidence from
Electric Vehicle Incentives,” American Economic Journal: Economic Policy, forthcoming.
Stavins, Robert, “Transaction Costs and Tradeable Permits,” Journal of Environmental Economics and
Management, 1995, 29 (2), 133–148.
Steck, Andrew L., “Industry Dynamics with Social Learning: Evidence from Hydraulic Fracturing,” 2018.
working paper.
Sweeney, Richard L., “Environmental Regulation, Imperfect Competition and Market Spillovers: The Im-
pact of the 1990 Clean Air Act Amendments on the US Oil Refining Industry,” 2015. working paper.
120
Timmins, Christopher, “Measuring the dynamic efficiency costs of regulators’ preferences: Municipal
water utilities in the Arid West,” Econometrica, 2002, 70 (2), 603–629.
Torchiana, Adrian L, Ted Rosenbaum, Paul T Scott, and Eduardo Souza-Rodrigues, “Improving Esti-
mates of Transitions from Satellite Data: A Hidden Markov Model Approach *,” 2020.
Toyama, Yuta, “Dynamic Incentives and Permit Market Equilibrium in Cap-and-Trade Regulation,” Mimeo,
2019.
Varian, Hal R., “A Model of Sales,” American Economic Review, 1980, 70, 651 – 659.
Vreugdenhil, Nicholas, “Booms, Busts, and Mismatch in Capital Markets: Evidence from the Offshore Oil
and Gas Industry,” 2020. working paper.
, “Incomplete Regulation and Spatial Equilibrium in the Offshore Oil and Gas Industry,” 2021. working
paper.
Ward, Melissa M. and Timothy W. Kelsey, “Natural Gas Lessors’ Experiences in Bradford and Tioga
Counties, 2010,” 2011. Marcellus Education Fact Sheet, PennState Extension.
Whitefoot, Kate S., Meredith L. Fowlie, and Steven J. Skerlos, “Influence of Acceleration Tradeoffs on
CO2 Emissions and Costs of Fuel Economy and Greenhouse Gas Regulations,” Environmental Science
and Technology, 2017, 51 (18), 10307 – 10315.
Wolak, Frank, “Designing Nonlinear Price Schedules for Urban Water Utilities to Balance Revenue and
Conservation Goals,” National Bureau of Economic Research, aug 2016.
Wolak, Frank A., “An Empirical Analysis of the Impact of Hedge Contracts on Bidding Behavior in a
Competitive Electricity Market *,” International Economic Journal, 2000, 14 (2), 1–39.
, Identification and Estimation of Cost Functions Using Observed Bid Data: An Application to Competi-
tive Electricity Markets, Cambridge University Press, 2003.
, “Quantifying the supply-side benefits from forward contracting in wholesale electricity markets,” Jour-
nal of Applied Econometrics, 2007, 22 (7), 1179–1209.
, “Measuring the competitiveness benefits of a transmission investment policy: The case of the Alberta
electricity market,” Energy Policy, 2015.
Wolak, Frank A, “Long-Term Resource Adequacy in Wholesale Electricity Markets With Significant In-
termittent Renewables,” 2021.
Wolfram, Catherine D., “Strategic Bidding in a Multiunit Auction: An Empirical Analysis of Bids to
Supply Electricity in England and Wales,” The RAND Journal of Economics, jan 1998, 29 (4), 703.
, “Measuring duopoly power in the British electricity spot market,” American Economic Review, 1999, 89
(4), 805–826.
121
Zou, Eric Yongchen, “Unwatched pollution: The effect of intermittent monitoring on air quality,” American
Economic Review, jul 2021, 111 (7), 2101–2126.
122